Authors: Mar\'ia Grandury
Abstract: We are 600 million Spanish speakers. We launched the #Somos600M Project because the diversity of the languages from LATAM, the Caribbean and Spain needs to be represented in Artificial Intelligence (AI) systems. Despite being the 7.5% of the world population, there is no open dataset to instruction-tune large language models (LLMs), nor a leaderboard to evaluate and compare them. In this paper, we present how we have created as an international open-source community the first versions of the instruction and evaluation datasets, indispensable resources for the advancement of Natural Language Processing (NLP) in our languages.
Authors: Fabi\'an Villena, Claudia V\'eliz, Rosario Garc\'ia-Huidobro, Sebasti\'an Aguayo
Abstract: Artificial intelligence (AI) has become a commodity for people because of the advent of generative AI (GenAI) models that bridge the usability gap of AI by providing a natural language interface to interact with complex models. These GenAI models range from text generation - such as two-way chat systems - to the generation of image or video from textual descriptions input by a user. These advancements in AI have impacted Dentistry in multiple aspects. In dental education, the student now has the opportunity to solve a plethora of questions by only prompting a GenAI model and have the answer in a matter of seconds. GenAI models can help us deliver better patient healthcare by helping practitioners gather knowledge quickly and efficiently. Finally, GenAI can also be used in dental research, where the applications range from new drug discovery to assistance in academic writing. In this review, we first define GenAI models and describe their multiple generation modalities; then, we explain and discuss their current and potential applications in Dentistry; and finally, we describe the challenges these new technologies impose in our area.
Authors: Ciprian Chelba, Johan Schalkwyk
Abstract: We present a modular approach to building cascade speech translation (AST) models that guarantees that the resulting model performs no worse than the 1-best cascade baseline while preserving state-of-the-art speech recognition (ASR) and text translation (MT) performance for a given task. Our novel contribution is the use of an ``exporter'' layer that is trained under L2-loss to ensure a strong match between ASR embeddings and the MT token embeddings for the 1-best sequence. The ``exporter'' output embeddings are fed directly to the MT model in lieu of 1-best token embeddings, thus guaranteeing that the resulting model performs no worse than the 1-best cascade baseline, while allowing back-propagation gradient to flow from the MT model into the ASR components. The matched-embeddings cascade architecture provide a significant improvement over its 1-best counterpart in scenarios where incremental training of the MT model is not an option and yet we seek to improve quality by leveraging (speech, transcription, translated transcription) data provided with the AST task. The gain disappears when the MT model is incrementally trained on the parallel text data available with the AST task. The approach holds promise for other scenarios that seek to couple ASR encoders and immutable text models, such at large language models (LLM).
Authors: Nikita Andreev, Alexander Shirnin, Vladislav Mikhailov, Ekaterina Artemova
Abstract: This paper presents Papilusion, an AI-generated scientific text detector developed within the DAGPap24 shared task on detecting automatically generated scientific papers. We propose an ensemble-based approach and conduct ablation studies to analyze the effect of the detector configurations on the performance. Papilusion is ranked 6th on the leaderboard, and we improve our performance after the competition ended, achieving 99.46 (+9.63) of the F1-score on the official test set.
Authors: An Quang Tang, Xiuzhen Zhang, Minh Ngoc Dinh
Abstract: This paper presents our proposed approach to the Discharge Me! shared task, collocated with the 23th Workshop on Biomedical Natural Language Processing (BioNLP). In this work, we develop an LLM-based framework for solving the Discharge Summary Documentation (DSD) task, i.e., generating the two critical target sections `Brief Hospital Course' and `Discharge Instructions' in the discharge summary. By streamlining the recent instruction-finetuning process on LLMs, we explore several prompting strategies for optimally adapting LLMs to specific generation task of DSD. Experimental results show that providing a clear output structure, complimented by a set of comprehensive Chain-of-Thoughts (CoT) questions, effectively improves the model's reasoning capability, and thereby, enhancing the structural correctness and faithfulness of clinical information in the generated text. Source code is available at: https://github.com/antangrocket1312/Discharge_LLM
Authors: Weisi Liu, Zhe He, Xiaolei Huang
Abstract: Time roots in applying language models for biomedical applications: models are trained on historical data and will be deployed for new or future data, which may vary from training data. While increasing biomedical tasks have employed state-of-the-art language models, there are very few studies have examined temporal effects on biomedical models when data usually shifts across development and deployment. This study fills the gap by statistically probing relations between language model performance and data shifts across three biomedical tasks. We deploy diverse metrics to evaluate model performance, distance methods to measure data drifts, and statistical methods to quantify temporal effects on biomedical language models. Our study shows that time matters for deploying biomedical language models, while the degree of performance degradation varies by biomedical tasks and statistical quantification approaches. We believe this study can establish a solid benchmark to evaluate and assess temporal effects on deploying biomedical language models.
Authors: Xihui Lin, Yunan Zhang, Suyu Ge, Barun Patra, Vishrav Chaudhary, Xia Song
Abstract: Existing LLM training and inference frameworks struggle in boosting efficiency with sparsity while maintaining the integrity of context and model architecture. Inspired by the sharding concept in database and the fact that attention parallelizes over heads on accelerators, we propose Sparsely-Sharded (S2) Attention, an attention algorithm that allocates heterogeneous context partitions for different attention heads to divide and conquer. S2-Attention enforces each attention head to only attend to a partition of contexts following a strided sparsity pattern, while the full context is preserved as the union of all the shards. As attention heads are processed in separate thread blocks, the context reduction for each head can thus produce end-to-end speed-up and memory reduction. At inference, LLMs trained with S2-Attention can then take the KV cache reduction as free meals with guaranteed model quality preserve. In experiments, we show S2-Attentioncan provide as much as (1) 25.3X wall-clock attention speed-up over FlashAttention-2, resulting in 6X reduction in end-to-end training time and 10X inference latency, (2) on-par model training quality compared to default attention, (3)perfect needle retrieval accuracy over 32K context window. On top of the algorithm, we build DKernel, an LLM training and inference kernel library that allows users to customize sparsity patterns for their own models. We open-sourced DKerneland make it compatible with Megatron, Pytorch, and vLLM.
Authors: Lynnette Hui Xian Ng, Iain Cruickshank, Roy Ka-Wei Lee
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in executing tasks based on natural language queries. However, these models, trained on curated datasets, inherently embody biases ranging from racial to national and gender biases. It remains uncertain whether these biases impact the performance of LLMs for certain tasks. In this study, we investigate the political biases of LLMs within the stance classification task, specifically examining whether these models exhibit a tendency to more accurately classify politically-charged stances. Utilizing three datasets, seven LLMs, and four distinct prompting schemes, we analyze the performance of LLMs on politically oriented statements and targets. Our findings reveal a statistically significant difference in the performance of LLMs across various politically oriented stance classification tasks. Furthermore, we observe that this difference primarily manifests at the dataset level, with models and prompting schemes showing statistically similar performances across different stance classification datasets. Lastly, we observe that when there is greater ambiguity in the target the statement is directed towards, LLMs have poorer stance classification accuracy.
Authors: Hao Shen, Zihan Li, Minqiang Yang, Minghui Ni, Yongfeng Tao, Zhengyang Yu, Weihao Zheng, Chen Xu, Bin Hu
Abstract: In contemporary society, the issue of psychological health has become increasingly prominent, characterized by the diversification, complexity, and universality of mental disorders. Cognitive Behavioral Therapy (CBT), currently the most influential and clinically effective psychological treatment method with no side effects, has limited coverage and poor quality in most countries. In recent years, researches on the recognition and intervention of emotional disorders using large language models (LLMs) have been validated, providing new possibilities for psychological assistance therapy. However, are LLMs truly possible to conduct cognitive behavioral therapy? Many concerns have been raised by mental health experts regarding the use of LLMs for therapy. Seeking to answer this question, we collected real CBT corpus from online video websites, designed and conducted a targeted automatic evaluation framework involving the evaluation of emotion tendency of generated text, structured dialogue pattern and proactive inquiry ability. For emotion tendency, we calculate the emotion tendency score of the CBT dialogue text generated by each model. For structured dialogue pattern, we use a diverse range of automatic evaluation metrics to compare speaking style, the ability to maintain consistency of topic and the use of technology in CBT between different models . As for inquiring to guide the patient, we utilize PQA (Proactive Questioning Ability) metric. We also evaluated the CBT ability of the LLM after integrating a CBT knowledge base to explore the help of introducing additional knowledge to enhance the model's CBT counseling ability. Four LLM variants with excellent performance on natural language processing are evaluated, and the experimental result shows the great potential of LLMs in psychological counseling realm, especially after combining with other technological means.
Authors: Xiaohan Fang, Chaozhuo Li, Yi Zhao, Qian Zang, Litian Zhang, Jiquan Peng, Xi Zhang, Jibing Gong
Abstract: Knowledge Graph Alignment (KGA) aims to integrate knowledge from multiple sources to address the limitations of individual Knowledge Graphs (KGs) in terms of coverage and depth. However, current KGA models fall short in achieving a ``complete'' knowledge graph alignment. Existing models primarily emphasize the linkage of cross-graph entities but overlook aligning relations across KGs, thereby providing only a partial solution to KGA. The semantic correlations embedded in relations are largely overlooked, potentially restricting a comprehensive understanding of cross-KG signals. In this paper, we propose to conceptualize relation alignment as an independent task and conduct KGA by decomposing it into two distinct but highly correlated sub-tasks: entity alignment and relation alignment. To capture the mutually reinforcing correlations between these objectives, we propose a novel Expectation-Maximization-based model, EREM, which iteratively optimizes both sub-tasks. Experimental results on real-world datasets demonstrate that EREM consistently outperforms state-of-the-art models in both entity alignment and relation alignment tasks.
Authors: Hyundong Cho, Thamme Gowda, Yuyang Huang, Zixun Lu, Tianli Tong, Jonathan May
Abstract: Following the rapid progress in natural language processing (NLP) models, language models are applied to increasingly more complex interactive tasks such as negotiations and conversation moderations. Having human evaluators directly interact with these NLP models is essential for adequately evaluating the performance on such interactive tasks. We develop BotEval, an easily customizable, open-source, evaluation toolkit that focuses on enabling human-bot interactions as part of the evaluation process, as opposed to human evaluators making judgements for a static input. BotEval balances flexibility for customization and user-friendliness by providing templates for common use cases that span various degrees of complexity and built-in compatibility with popular crowdsourcing platforms. We showcase the numerous useful features of BotEval through a study that evaluates the performance of various chatbots on their effectiveness for conversational moderation and discuss how BotEval differs from other annotation tools.
Authors: Mattia Opper, N. Siddharth
Abstract: We present Banyan, an improved model to learn semantic representations by inducing explicit structure over data. In contrast to prior approaches using structure spanning single sentences, Banyan learns by resolving multiple constituent structures into a shared one explicitly incorporating global context. Combined with an improved message-passing scheme inspired by Griffin, Banyan learns significantly better representations, avoids spurious false negatives with contrastive learning, and drastically improves memory efficiency in such explicit-structured models. Using the Self-StrAE framework, we show that Banyan (a) outperforms baselines using sentential structure across various settings (b) matches or outperforms unstructured baselines like GloVe (+augmentations) and a RoBERTa medium (+simcse) pre-trained on 100M tokens, despite having just a handful of (non-embedding) parameters, and (c) also learns effective representations across several low resource (Asian and African) languages as measured on SemRel tasks.
Authors: Jing Huang, Diyi Yang, Christopher Potts
Abstract: Large Language Models (LLMs) frequently memorize long sequences verbatim, often with serious legal and privacy implications. Much prior work has studied such verbatim memorization using observational data. To complement such work, we develop a framework to study verbatim memorization in a controlled setting by continuing pre-training from Pythia checkpoints with injected sequences. We find that (1) non-trivial amounts of repetition are necessary for verbatim memorization to happen; (2) later (and presumably better) checkpoints are more likely to verbatim memorize sequences, even for out-of-distribution sequences; (3) the generation of memorized sequences is triggered by distributed model states that encode high-level features and makes important use of general language modeling capabilities. Guided by these insights, we develop stress tests to evaluate unlearning methods and find they often fail to remove the verbatim memorized information, while also degrading the LM. Overall, these findings challenge the hypothesis that verbatim memorization stems from specific model weights or mechanisms. Rather, verbatim memorization is intertwined with the LM's general capabilities and thus will be very difficult to isolate and suppress without degrading model quality.
Authors: Jinming Zhao, Vineel Pratap, Michael Auli
Abstract: Despite rapid progress in increasing the language coverage of automatic speech recognition, the field is still far from covering all languages with a known writing script. Recent work showed promising results with a zero-shot approach requiring only a small amount of text data, however, accuracy heavily depends on the quality of the used phonemizer which is often weak for unseen languages. In this paper, we present MMS Zero-shot a conceptually simpler approach based on romanization and an acoustic model trained on data in 1,078 different languages or three orders of magnitude more than prior art. MMS Zero-shot reduces the average character error rate by a relative 46% over 100 unseen languages compared to the best previous work. Moreover, the error rate of our approach is only 2.5x higher compared to in-domain supervised baselines, while our approach uses no labeled data for the evaluation languages at all.
Authors: Ruoyu Hu, Foaad Khosmood, Abbas Edalat
Abstract: In this work, we introduce several schemes to leverage description-augmented embedding similarity for dataless intent classification using current state-of-the-art (SOTA) text embedding models. We report results of our methods on four commonly used intent classification datasets and compare against previous works of a similar nature. Our work shows promising results for dataless classification scaling to a large number of unseen intents. We show competitive results and significant improvements (+6.12\% Avg.) over strong zero-shot baselines, all without training on labelled or task-specific data. Furthermore, we provide qualitative error analysis of the shortfalls of this methodology to help guide future research in this area.
Authors: Zden\v{e}k Kasner, Ond\v{r}ej Pl\'atek, Patr\'icia Schmidtov\'a, Simone Balloccu, Ond\v{r}ej Du\v{s}ek
Abstract: We present factgenie: a framework for annotating and visualizing word spans in textual model outputs. Annotations can capture various span-based phenomena such as semantic inaccuracies or irrelevant text. With factgenie, the annotations can be collected both from human crowdworkers and large language models. Our framework consists of a web interface for data visualization and gathering text annotations, powered by an easily extensible codebase.
Authors: Jiwon Suh, Injae Na, Woohwan Jung
Abstract: End-to-end automatic speech recognition (E2E ASR) systems have significantly improved speech recognition through training on extensive datasets. Despite these advancements, they still struggle to accurately recognize domain specific words, such as proper nouns and technical terminologies. To address this problem, we propose a method to utilize the state-of-the-art Whisper without modifying its architecture, preserving its generalization performance while enabling it to leverage descriptions effectively. Moreover, we propose two additional training techniques to improve the domain specific ASR: decoder fine-tuning, and context perturbation. We also propose a method to use a Large Language Model (LLM) to generate descriptions with simple metadata, when descriptions are unavailable. Our experiments demonstrate that proposed methods notably enhance domain-specific ASR accuracy on real-life datasets, with LLM-generated descriptions outperforming human-crafted ones in effectiveness.
Authors: Daniel Atzberger, Tim Cech, Willy Scheibel, J\"urgen D\"ollner, Michael Behrisch, Tobias Schreck
Abstract: The semantic similarity between documents of a text corpus can be visualized using map-like metaphors based on two-dimensional scatterplot layouts. These layouts result from a dimensionality reduction on the document-term matrix or a representation within a latent embedding, including topic models. Thereby, the resulting layout depends on the input data and hyperparameters of the dimensionality reduction and is therefore affected by changes in them. Furthermore, the resulting layout is affected by changes in the input data and hyperparameters of the dimensionality reduction. However, such changes to the layout require additional cognitive efforts from the user. In this work, we present a sensitivity study that analyzes the stability of these layouts concerning (1) changes in the text corpora, (2) changes in the hyperparameter, and (3) randomness in the initialization. Our approach has two stages: data measurement and data analysis. First, we derived layouts for the combination of three text corpora and six text embeddings and a grid-search-inspired hyperparameter selection of the dimensionality reductions. Afterward, we quantified the similarity of the layouts through ten metrics, concerning local and global structures and class separation. Second, we analyzed the resulting 42817 tabular data points in a descriptive statistical analysis. From this, we derived guidelines for informed decisions on the layout algorithm and highlight specific hyperparameter settings. We provide our implementation as a Git repository at https://github.com/hpicgs/Topic-Models-and-Dimensionality-Reduction-Sensitivity-Study and results as Zenodo archive at https://doi.org/10.5281/zenodo.12772898.
URLs: https://github.com/hpicgs/Topic-Models-and-Dimensionality-Reduction-Sensitivity-Study, https://doi.org/10.5281/zenodo.12772898.
Authors: Danqing Hu, Bing Liu, Xiaofeng Zhu, Nan Wu
Abstract: Lymph node metastasis (LNM) is a crucial factor in determining the initial treatment for patients with lung cancer, yet accurate preoperative diagnosis of LNM remains challenging. Recently, large language models (LLMs) have garnered significant attention due to their remarkable text generation capabilities. Leveraging the extensive medical knowledge learned from vast corpora, LLMs can estimate probabilities for clinical problems, though their performance has historically been inferior to data-driven machine learning models. In this paper, we propose a novel ensemble method that combines the medical knowledge acquired by LLMs with the latent patterns identified by machine learning models to enhance LNM prediction performance. Initially, we developed machine learning models using patient data. We then designed a prompt template to integrate the patient data with the predicted probability from the machine learning model. Subsequently, we instructed GPT-4o, the most advanced LLM developed by OpenAI, to estimate the likelihood of LNM based on patient data and then adjust the estimate using the machine learning output. Finally, we collected three outputs from the GPT-4o using the same prompt and ensembled these results as the final prediction. Using the proposed method, our models achieved an AUC value of 0.765 and an AP value of 0.415 for LNM prediction, significantly improving predictive performance compared to baseline machine learning models. The experimental results indicate that GPT-4o can effectively leverage its medical knowledge and the probabilities predicted by machine learning models to achieve more accurate LNM predictions. These findings demonstrate that LLMs can perform well in clinical risk prediction tasks, offering a new paradigm for integrating medical knowledge and patient data in clinical predictions.
Authors: Anna Bavaresco, Marianne de Heer Kloots, Sandro Pezzelle, Raquel Fern\'andez
Abstract: Representations from deep neural networks (DNNs) have proven remarkably predictive of neural activity involved in both visual and linguistic processing. Despite these successes, most studies to date concern unimodal DNNs, encoding either visual or textual input but not both. Yet, there is growing evidence that human meaning representations integrate linguistic and sensory-motor information. Here we investigate whether the integration of multimodal information operated by current vision-and-language DNN models (VLMs) leads to representations that are more aligned with human brain activity than those obtained by language-only and vision-only DNNs. We focus on fMRI responses recorded while participants read concept words in the context of either a full sentence or an accompanying picture. Our results reveal that VLM representations correlate more strongly than language- and vision-only DNNs with activations in brain areas functionally related to language processing. A comparison between different types of visuo-linguistic architectures shows that recent generative VLMs tend to be less brain-aligned than previous architectures with lower performance on downstream applications. Moreover, through an additional analysis comparing brain vs. behavioural alignment across multiple VLMs, we show that -- with one remarkable exception -- representations that strongly align with behavioural judgments do not correlate highly with brain responses. This indicates that brain similarity does not go hand in hand with behavioural similarity, and vice versa.
Authors: Shutong Jia, Biwei Cao, Qingqing Gao, Jiuxin Cao, Bo Liu
Abstract: Differing from sentiment transfer, positive reframing seeks to substitute negative perspectives with positive expressions while preserving the original meaning. With the emergence of pre-trained language models (PLMs), it is possible to achieve acceptable results by fine-tuning PLMs. Nevertheless, generating fluent, diverse and task-constrained reframing text remains a significant challenge. To tackle this issue, a \textbf{m}ulti-\textbf{s}trategy \textbf{o}ptimization \textbf{f}ramework (MSOF) is proposed in this paper. Starting from the objective of positive reframing, we first design positive sentiment reward and content preservation reward to encourage the model to transform the negative expressions of the original text while ensuring the integrity and consistency of the semantics. Then, different decoding optimization approaches are introduced to improve the quality of text generation. Finally, based on the modeling formula of positive reframing, we propose a multi-dimensional re-ranking method that further selects candidate sentences from three dimensions: strategy consistency, text similarity and fluency. Extensive experiments on two Seq2Seq PLMs, BART and T5, demonstrate our framework achieves significant improvements on unconstrained and controlled positive reframing tasks.
Authors: Tom Kouwenhoven, Max Peeperkorn, Bram van Dijk, Tessa Verhoef
Abstract: Natural language has the universal properties of being compositional and grounded in reality. The emergence of linguistic properties is often investigated through simulations of emergent communication in referential games. However, these experiments have yielded mixed results compared to similar experiments addressing linguistic properties of human language. Here we address representational alignment as a potential contributing factor to these results. Specifically, we assess the representational alignment between agent image representations and between agent representations and input images. Doing so, we confirm that the emergent language does not appear to encode human-like conceptual visual features, since agent image representations drift away from inputs whilst inter-agent alignment increases. We moreover identify a strong relationship between inter-agent alignment and topographic similarity, a common metric for compositionality, and address its consequences. To address these issues, we introduce an alignment penalty that prevents representational drift but interestingly does not improve performance on a compositional discrimination task. Together, our findings emphasise the key role representational alignment plays in simulations of language emergence.
Authors: Tessa Verhoef, Kiana Shahrasbi, Tom Kouwenhoven
Abstract: Humans have clear cross-modal preferences when matching certain novel words to visual shapes. Evidence suggests that these preferences play a prominent role in our linguistic processing, language learning, and the origins of signal-meaning mappings. With the rise of multimodal models in AI, such as vision- and-language (VLM) models, it becomes increasingly important to uncover the kinds of visio-linguistic associations these models encode and whether they align with human representations. Informed by experiments with humans, we probe and compare four VLMs for a well-known human cross-modal preference, the bouba-kiki effect. We do not find conclusive evidence for this effect but suggest that results may depend on features of the models, such as architecture design, model size, and training details. Our findings inform discussions on the origins of the bouba-kiki effect in human cognition and future developments of VLMs that align well with human cross-modal associations.
Authors: Nick Rossenbach, Benedikt Hilmes, Ralf Schl\"uter
Abstract: In this work we evaluate the utility of synthetic data for training automatic speech recognition (ASR). We use the ASR training data to train a text-to-speech (TTS) system similar to FastSpeech-2. With this TTS we reproduce the original training data, training ASR systems solely on synthetic data. For ASR, we use three different architectures, attention-based encoder-decoder, hybrid deep neural network hidden Markov model and a Gaussian mixture hidden Markov model, showing the different sensitivity of the models to synthetic data generation. In order to extend previous work, we present a number of ablation studies on the effectiveness of synthetic vs. real training data for ASR. In particular we focus on how the gap between training on synthetic and real data changes by varying the speaker embedding or by scaling the model size. For the latter we show that the TTS models generalize well, even when training scores indicate overfitting.
Authors: Shi Luohe, Zhang Hongyi, Yao Yao, Li Zuchao, Zhao Hai
Abstract: Large Language Models (LLMs), epitomized by ChatGPT' s release in late 2022, have revolutionized various industries with their advanced language comprehension. However, their efficiency is challenged by the Transformer architecture' s struggle with handling long texts. KV-Cache has emerged as a pivotal solution to this issue, converting the time complexity of token generation from quadratic to linear, albeit with increased GPU memory overhead proportional to conversation length. With the development of the LLM community and academia, various KV-Cache compression methods have been proposed. In this review, we dissect the various properties of KV-Cache and elaborate on various methods currently used to optimize the KV-Cache space usage of LLMs. These methods span the pre-training phase, deployment phase, and inference phase, and we summarize the commonalities and differences among these methods. Additionally, we list some metrics for evaluating the long-text capabilities of large language models, from both efficiency and capability perspectives. Our review thus sheds light on the evolving landscape of LLM optimization, offering insights into future advancements in this dynamic field.
Authors: Henri Jamet, Yash Raj Shrestha, Michalis Vlachos
Abstract: We leverage generative large language models for language learning applications, focusing on estimating the difficulty of foreign language texts and simplifying them to lower difficulty levels. We frame both tasks as prediction problems and develop a difficulty classification model using labeled examples, transfer learning, and large language models, demonstrating superior accuracy compared to previous approaches. For simplification, we evaluate the trade-off between simplification quality and meaning preservation, comparing zero-shot and fine-tuned performances of large language models. We show that meaningful text simplifications can be obtained with limited fine-tuning. Our experiments are conducted on French texts, but our methods are language-agnostic and directly applicable to other foreign languages.
Authors: Christopher Clarke, Yuzhao Heng, Lingjia Tang, Jason Mars
Abstract: The recent emergence of Large Language Models (LLMs) has heralded a new era of human-AI interaction. These sophisticated models, exemplified by Chat-GPT and its successors, have exhibited remarkable capabilities in language understanding. However, as these LLMs have undergone exponential growth, a crucial dimension that remains understudied is the personalization of these models. Large foundation models such as GPT-3 etc. focus on creating a universal model that serves a broad range of tasks and users. This approach emphasizes the model's generalization capabilities, treating users as a collective rather than as distinct individuals. While practical for many common applications, this one-size-fits-all approach often fails to address the rich tapestry of human diversity and individual needs. To explore this issue we introduce the PEFT-U Benchmark: a new dataset for building and evaluating NLP models for user personalization. \datasetname{} consists of a series of user-centered tasks containing diverse and individualized expressions where the preferences of users can potentially differ for the same input. Using PEFT-U, we explore the challenge of efficiently personalizing LLMs to accommodate user-specific preferences in the context of diverse user-centered tasks.
Authors: Vivi Nastase, Paola Merlo
Abstract: Analyses of transformer-based models have shown that they encode a variety of linguistic information from their textual input. While these analyses have shed a light on the relation between linguistic information on one side, and internal architecture and parameters on the other, a question remains unanswered: how is this linguistic information reflected in sentence embeddings? Using datasets consisting of sentences with known structure, we test to what degree information about chunks (in particular noun, verb or prepositional phrases), such as grammatical number, or semantic role, can be localized in sentence embeddings. Our results show that such information is not distributed over the entire sentence embedding, but rather it is encoded in specific regions. Understanding how the information from an input text is compressed into sentence embeddings helps understand current transformer models and help build future explainable neural models.
Authors: Fakhraddin Alwajih, Gagan Bhatia, Muhammad Abdul-Mageed
Abstract: Recent advancements have significantly enhanced the capabilities of Multimodal Large Language Models (MLLMs) in generating and understanding image-to-text content. Despite these successes, progress is predominantly limited to English due to the scarcity of high quality multimodal resources in other languages. This limitation impedes the development of competitive models in languages such as Arabic. To alleviate this situation, we introduce an efficient Arabic multimodal assistant, dubbed Dallah, that utilizes an advanced language model based on LLaMA-2 to facilitate multimodal interactions. Dallah demonstrates state-of-the-art performance in Arabic MLLMs. Through fine-tuning six Arabic dialects, Dallah showcases its capability to handle complex dialectal interactions incorporating both textual and visual elements. The model excels in two benchmark tests: one evaluating its performance on Modern Standard Arabic (MSA) and another specifically designed to assess dialectal responses. Beyond its robust performance in multimodal interaction tasks, Dallah has the potential to pave the way for further development of dialect-aware Arabic MLLMs.
Authors: Wajdi Zaghouani (Northwestern University in Qatar), Mustafa Jarrar (Birzeit University), Nizar Habash (New York University Abu Dhabi), Houda Bouamor (Carnegie Mellon University Qatar), Imed Zitouni (Google), Mona Diab (Carnegie Mellon University), Samhaa R. El-Beltagy (Newgiza University), Muhammed AbuOdeh (New York University Abu Dhabi)
Abstract: We present an overview of the FIGNEWS shared task, organized as part of the ArabicNLP 2024 conference co-located with ACL 2024. The shared task addresses bias and propaganda annotation in multilingual news posts. We focus on the early days of the Israel War on Gaza as a case study. The task aims to foster collaboration in developing annotation guidelines for subjective tasks by creating frameworks for analyzing diverse narratives highlighting potential bias and propaganda. In a spirit of fostering and encouraging diversity, we address the problem from a multilingual perspective, namely within five languages: English, French, Arabic, Hebrew, and Hindi. A total of 17 teams participated in two annotation subtasks: bias (16 teams) and propaganda (6 teams). The teams competed in four evaluation tracks: guidelines development, annotation quality, annotation quantity, and consistency. Collectively, the teams produced 129,800 data points. Key findings and implications for the field are discussed.
Authors: Tianduo Wang, Shichen Li, Wei Lu
Abstract: Effective training of language models (LMs) for mathematical reasoning tasks demands high-quality supervised fine-tuning data. Besides obtaining annotations from human experts, a common alternative is sampling from larger and more powerful LMs. However, this knowledge distillation approach can be costly and unstable, particularly when relying on closed-source, proprietary LMs like GPT-4, whose behaviors are often unpredictable. In this work, we demonstrate that the reasoning abilities of small-scale LMs can be enhanced through self-training, a process where models learn from their own outputs. We also show that the conventional self-training can be further augmented by a preference learning algorithm called Direct Preference Optimization (DPO). By integrating DPO into self-training, we leverage preference data to guide LMs towards more accurate and diverse chain-of-thought reasoning. We evaluate our method across various mathematical reasoning tasks using different base models. Our experiments show that this approach not only improves LMs' reasoning performance but also offers a more cost-effective and scalable solution compared to relying on large proprietary LMs.
Authors: Xingyao Wang, Boxuan Li, Yufan Song, Frank F. Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang H. Tran, Fuqiang Li, Ren Ma, Mingzhang Zheng, Bill Qian, Yanjun Shao, Niklas Muennighoff, Yizhe Zhang, Binyuan Hui, Junyang Lin, Robert Brennan, Hao Peng, Heng Ji, Graham Neubig
Abstract: Software is one of the most powerful tools that we humans have at our disposal; it allows a skilled programmer to interact with the world in complex and profound ways. At the same time, thanks to improvements in large language models (LLMs), there has also been a rapid development in AI agents that interact with and affect change in their surrounding environments. In this paper, we introduce OpenDevin, a platform for the development of powerful and flexible AI agents that interact with the world in similar ways to those of a human developer: by writing code, interacting with a command line, and browsing the web. We describe how the platform allows for the implementation of new agents, safe interaction with sandboxed environments for code execution, coordination between multiple agents, and incorporation of evaluation benchmarks. Based on our currently incorporated benchmarks, we perform an evaluation of agents over 15 challenging tasks, including software engineering (e.g., SWE-Bench) and web browsing (e.g., WebArena), among others. Released under the permissive MIT license, OpenDevin is a community project spanning academia and industry with more than 1.3K contributions from over 160 contributors and will improve going forward.
Authors: Feng Chen, Manas Satish Bedmutha, Ray-Yuan Chung, Janice Sabin, Wanda Pratt, Brian R. Wood, Nadir Weibel, Andrea L. Hartzler, Trevor Cohen
Abstract: Implicit bias can impede patient-provider interactions and lead to inequities in care. Raising awareness is key to reducing such bias, but its manifestations in the social dynamics of patient-provider communication are difficult to detect. In this study, we used automated speech recognition (ASR) and natural language processing (NLP) to identify social signals in patient-provider interactions. We built an automated pipeline to predict social signals from audio recordings of 782 primary care visits that achieved 90.1% average accuracy across codes, and exhibited fairness in its predictions for white and non-white patients. Applying this pipeline, we identified statistically significant differences in provider communication behavior toward white versus non-white patients. In particular, providers expressed more patient-centered behaviors towards white patients including more warmth, engagement, and attentiveness. Our study underscores the potential of automated tools in identifying subtle communication signals that may be linked with bias and impact healthcare quality and equity.
Authors: Clara Ziche, Giovanni Apruzzese
Abstract: We investigate the potential of using Large Language Models (LLM) to support process model creation in organizational contexts. Specifically, we carry out a case study wherein we develop and test an LLM-based chatbot, PRODIGY (PROcess moDellIng Guidance for You), in a multinational company, the Hilti Group. We are particularly interested in understanding how LLM can aid (human) modellers in creating process flow diagrams. To this purpose, we first conduct a preliminary user study (n=10) with professional process modellers from Hilti, inquiring for various pain-points they encounter in their daily routines. Then, we use their responses to design and implement PRODIGY. Finally, we evaluate PRODIGY by letting our user study's participants use PRODIGY, and then ask for their opinion on the pros and cons of PRODIGY. We coalesce our results in actionable takeaways. Through our research, we showcase the first practical application of LLM for process modelling in the real world, shedding light on how industries can leverage LLM to enhance their Business Process Management activities.
Authors: Kristian Gonz\'alez Barman, Simon Lohse, Henk de Regt
Abstract: We argue for the epistemic and ethical advantages of pluralism in Reinforcement Learning from Human Feedback (RLHF) in the context of Large Language Models (LLM). Drawing on social epistemology and pluralist philosophy of science, we suggest ways in which RHLF can be made more responsive to human needs and how we can address challenges along the way. The paper concludes with an agenda for change, i.e. concrete, actionable steps to improve LLM development.
Authors: Keane Ong, Rui Mao, Ranjan Satapathy, Ricardo Shirota Filho, Erik Cambria, Johan Sulaeman, Gianmarco Mengaldo
Abstract: Sustainability commonly refers to entities, such as individuals, companies, and institutions, having a non-detrimental (or even positive) impact on the environment, society, and the economy. With sustainability becoming a synonym of acceptable and legitimate behaviour, it is being increasingly demanded and regulated. Several frameworks and standards have been proposed to measure the sustainability impact of corporations, including United Nations' sustainable development goals and the recently introduced global sustainability reporting framework, amongst others. However, the concept of corporate sustainability is complex due to the diverse and intricate nature of firm operations (i.e. geography, size, business activities, interlinks with other stakeholders). As a result, corporate sustainability assessments are plagued by subjectivity both within data that reflect corporate sustainability efforts (i.e. corporate sustainability disclosures) and the analysts evaluating them. This subjectivity can be distilled into distinct challenges, such as incompleteness, ambiguity, unreliability and sophistication on the data dimension, as well as limited resources and potential bias on the analyst dimension. Put together, subjectivity hinders effective cost attribution to entities non-compliant with prevailing sustainability expectations, potentially rendering sustainability efforts and its associated regulations futile. To this end, we argue that Explainable Natural Language Processing (XNLP) can significantly enhance corporate sustainability analysis. Specifically, linguistic understanding algorithms (lexical, semantic, syntactic), integrated with XAI capabilities (interpretability, explainability, faithfulness), can bridge gaps in analyst resources and mitigate subjectivity problems within data.
Authors: Harshil Darji, Jelena Mitrovi\'c, Michael Granitzer
Abstract: The process of annotating data within the legal sector is filled with distinct challenges that differ from other fields, primarily due to the inherent complexities of legal language and documentation. The initial task usually involves selecting an appropriate raw dataset that captures the intricate aspects of legal texts. Following this, extracting text becomes a complicated task, as legal documents often have complex structures, footnotes, references, and unique terminology. The importance of data cleaning is magnified in this context, ensuring that redundant information is eliminated while maintaining crucial legal details and context. Creating comprehensive yet straightforward annotation guidelines is imperative, as these guidelines serve as the road map for maintaining uniformity and addressing the subtle nuances of legal terminology. Another critical aspect is the involvement of legal professionals in the annotation process. Their expertise is valuable in ensuring that the data not only remains contextually accurate but also adheres to prevailing legal standards and interpretations. This paper provides an expanded view of these challenges and aims to offer a foundational understanding and guidance for researchers and professionals engaged in legal data annotation projects. In addition, we provide links to our created and fine-tuned datasets and language models. These resources are outcomes of our discussed projects and solutions to challenges faced while working on them.
Authors: Aki H\"arm\"a, Bert den Brinker, Ulf Grossekathofer, Okke Ouweltjes, Srikanth Nallanthighal, Sidharth Abrol, Vibhu Sharma
Abstract: Recent years has witnessed an increase in technologies that use speech for the sensing of the health of the talker. This survey paper proposes a general taxonomy of the technologies and a broad overview of current progress and challenges. Vocal biomarkers are often secondary measures that are approximating a signal of another sensor or identifying an underlying mental, cognitive, or physiological state. Their measurement involve disturbances and uncertainties that may be considered as noise sources and the biomarkers are coarsely qualified in terms of the various sources of noise involved in their determination. While in some proposed biomarkers the error levels seem high, there are vocal biomarkers where the errors are expected to be low and thus are more likely to qualify as candidates for adoption in healthcare applications.
Authors: Alina Landowska, Maciej Skorski, Krzysztof Rajda
Abstract: People worldwide are currently confronted with a number of technological challenges, which act as a potent source of uncertainty. The uncertainty arising from the volatility and unpredictability of technology (such as AI) and its potential consequences is widely discussed on social media. This study uses BERTopic modelling along with sentiment and emotion analysis on 1.5 million tweets from 2021 to 2023 to identify anticipated tech-driven futures and capture the emotions communicated by 400 key opinion leaders (KOLs). Findings indicate positive sentiment significantly outweighs negative, with a prevailing dominance of positive anticipatory emotions. Specifically, the 'Hope' score is approximately 10.33\% higher than the median 'Anxiety' score. KOLs emphasize 'Optimism' and benefits over 'Pessimism' and challenges. The study emphasizes the important role KOLs play in shaping future visions through anticipatory discourse and emotional tone during times of technological uncertainty.
Authors: Hongwei Jin, George Papadimitriou, Krishnan Raghavan, Pawel Zuk, Prasanna Balaprakash, Cong Wang, Anirban Mandal, Ewa Deelman
Abstract: Anomaly detection in computational workflows is critical for ensuring system reliability and security. However, traditional rule-based methods struggle to detect novel anomalies. This paper leverages large language models (LLMs) for workflow anomaly detection by exploiting their ability to learn complex data patterns. Two approaches are investigated: 1) supervised fine-tuning (SFT), where pre-trained LLMs are fine-tuned on labeled data for sentence classification to identify anomalies, and 2) in-context learning (ICL) where prompts containing task descriptions and examples guide LLMs in few-shot anomaly detection without fine-tuning. The paper evaluates the performance, efficiency, generalization of SFT models, and explores zero-shot and few-shot ICL prompts and interpretability enhancement via chain-of-thought prompting. Experiments across multiple workflow datasets demonstrate the promising potential of LLMs for effective anomaly detection in complex executions.
Authors: Hyuk Namgoong, Jeesu Jung, Sangkeun Jung, Yoonhyung Roh
Abstract: Recent advancements in large language models have heavily relied on the large reward model from reinforcement learning from human feedback for fine-tuning. However, the use of a single reward model across various domains may not always be optimal, often requiring retraining from scratch when new domain data is introduced. To address these challenges, we explore the utilization of small language models operating in a domain-specific manner based on router mechanisms. Our three approaches are: 1) utilize mixture of experts to form a single reward model by modularizing an internal router and experts, 2) employing external router to select the appropriate reward model from multiple domain-specific models, and 3) the framework reduces parameter size by loading reward models and router adapters onto a single small language model using adapters. Experimental validation underscores the effectiveness of our approach, demonstrating performance comparable to baseline methods while also reducing the total parameter size.
Authors: Iustin Floroiu
Abstract: Given the high advances of large language models (LLM) it is of vital importance to study their behaviors and apply their utility in all kinds of scientific fields. Psychology has been, in recent years, poorly approached using novel computational tools. One of the reasons is the high complexity of the data required for a proper analysis. Moreover, psychology, with a focus on psychometry, has few datasets available for analysis and artificial intelligence usage. Because of these facts, this study introduces a synthethic database of short essays labeled based on the five factor model (FFM) of personality traits.
Authors: Felix Drinkall, Janet B. Pierrehumbert, Stefan Zohren
Abstract: Large Language Models (LLMs) have been shown to perform well for many downstream tasks. Transfer learning can enable LLMs to acquire skills that were not targeted during pre-training. In financial contexts, LLMs can sometimes beat well-established benchmarks. This paper investigates how well LLMs perform in the task of forecasting corporate credit ratings. We show that while LLMs are very good at encoding textual information, traditional methods are still very competitive when it comes to encoding numeric and multimodal data. For our task, current LLMs perform worse than a more traditional XGBoost architecture that combines fundamental and macroeconomic data with high-density text-based embedding features.
Authors: Nived Rajaraman, Marco Bondaschi, Kannan Ramchandran, Michael Gastpar, Ashok Vardhan Makkuva
Abstract: Attention-based transformers have been remarkably successful at modeling generative processes across various domains and modalities. In this paper, we study the behavior of transformers on data drawn from \kth Markov processes, where the conditional distribution of the next symbol in a sequence depends on the previous $k$ symbols observed. We observe a surprising phenomenon empirically which contradicts previous findings: when trained for sufficiently long, a transformer with a fixed depth and $1$ head per layer is able to achieve low test loss on sequences drawn from \kth Markov sources, even as $k$ grows. Furthermore, this low test loss is achieved by the transformer's ability to represent and learn the in-context conditional empirical distribution. On the theoretical side, our main result is that a transformer with a single head and three layers can represent the in-context conditional empirical distribution for \kth Markov sources, concurring with our empirical observations. Along the way, we prove that \textit{attention-only} transformers with $O(\log_2(k))$ layers can represent the in-context conditional empirical distribution by composing induction heads to track the previous $k$ symbols in the sequence. These results provide more insight into our current understanding of the mechanisms by which transformers learn to capture context, by understanding their behavior on Markov sources.
Authors: Zhiyuan Sun, Haochen Shi, Marc-Alexandre C\^ot\'e, Glen Berseth, Xingdi Yuan, Bang Liu
Abstract: While large language models (LLMs) have been increasingly deployed across tasks in language understanding and interactive decision-making, their impressive performance is largely due to the comprehensive and in-depth domain knowledge embedded within them. However, the extent of this knowledge can vary across different domains. Existing methods often assume that LLMs already possess such comprehensive and in-depth knowledge of their environment, overlooking potential gaps in their understanding of actual world dynamics. To address this gap, we introduce Discover, Verify, and Evolve (DiVE), a framework that discovers world dynamics from a small number of demonstrations, verifies the correctness of these dynamics, and evolves new, advanced dynamics tailored to the current situation. Through extensive evaluations, we analyze the impact of each component on performance and compare the automatically generated dynamics from DiVE with human-annotated world dynamics. Our results demonstrate that LLMs guided by DiVE can make better decisions, achieving rewards comparable to human players in the Crafter environment.
Authors: Seong-Gyun Leem, Daniel Fulford, Jukka-Pekka Onnela, David Gard, Carlos Busso
Abstract: Speech emotion recognition (SER) systems often struggle in real-world environments, where ambient noise severely degrades their performance. This paper explores a novel approach that exploits prior knowledge of testing environments to maximize SER performance under noisy conditions. To address this task, we propose a text-guided, environment-aware training where an SER model is trained with contaminated speech samples and their paired noise description. We use a pre-trained text encoder to extract the text-based environment embedding and then fuse it to a transformer-based SER model during training and inference. We demonstrate the effectiveness of our approach through our experiment with the MSP-Podcast corpus and real-world additive noise samples collected from the Freesound repository. Our experiment indicates that the text-based environment descriptions processed by a large language model (LLM) produce representations that improve the noise-robustness of the SER system. In addition, our proposed approach with an LLM yields better performance than our environment-agnostic baselines, especially in low signal-to-noise ratio (SNR) conditions. When testing at -5dB SNR level, our proposed method shows better performance than our best baseline model by 31.8 % (arousal), 23.5% (dominance), and 9.5% (valence).
Authors: Kaitao Chen, Mianxin Liu, Fang Yan, Lei Ma, Xiaoming Shi, Lilong Wang, Xiaosong Wang, Lifeng Zhu, Zhe Wang, Mu Zhou, Shaoting Zhang
Abstract: The advent of vision-language models fosters the interactive conversations between AI-enabled models and humans. Yet applying these models into clinics must deal with daunting challenges around large-scale training data, financial, and computational resources. Here we propose a cost-effective instruction learning framework for conversational pathology named as CLOVER. CLOVER only trains a lightweight module and uses instruction tuning while freezing the parameters of the large language model. Instead of using costly GPT-4, we propose well-designed prompts on GPT-3.5 for building generation-based instructions, emphasizing the utility of pathological knowledge derived from the Internet source. To augment the use of instructions, we construct a high-quality set of template-based instructions in the context of digital pathology. From two benchmark datasets, our findings reveal the strength of hybrid-form instructions in the visual question-answer in pathology. Extensive results show the cost-effectiveness of CLOVER in answering both open-ended and closed-ended questions, where CLOVER outperforms strong baselines that possess 37 times more training parameters and use instruction data generated from GPT-4. Through the instruction tuning, CLOVER exhibits robustness of few-shot learning in the external clinical dataset. These findings demonstrate that cost-effective modeling of CLOVER could accelerate the adoption of rapid conversational applications in the landscape of digital pathology.
Authors: Rita Frieske, Bertrand E. Shi
Abstract: ERIT is a novel multimodal dataset designed to facilitate research in a lightweight multimodal fusion. It contains text and image data collected from videos of elderly individuals reacting to various situations, as well as seven emotion labels for each data sample. Because of the use of labeled images of elderly users reacting emotionally, it is also facilitating research on emotion recognition in an underrepresented age group in machine learning visual emotion recognition. The dataset is validated through comprehensive experiments indicating its importance in neural multimodal fusion research.
Authors: Eunice Yiu, Maan Qraitem, Charlie Wong, Anisa Noor Majhi, Yutong Bai, Shiry Ginosar, Alison Gopnik, Kate Saenko
Abstract: This paper investigates visual analogical reasoning in large multimodal models (LMMs) compared to human adults and children. A "visual analogy" is an abstract rule inferred from one image and applied to another. While benchmarks exist for testing visual reasoning in LMMs, they require advanced skills and omit basic visual analogies that even young children can make. Inspired by developmental psychology, we propose a new benchmark of 1,400 visual transformations of everyday objects to test LMMs on visual analogical reasoning and compare them to children and adults. We structure the evaluation into three stages: identifying what changed (e.g., color, number, etc.), how it changed (e.g., added one object), and applying the rule to new scenarios. Our findings show that while models like GPT-4V, LLaVA-1.5, and MANTIS identify the "what" effectively, they struggle with quantifying the "how" and extrapolating this rule to new objects. In contrast, children and adults exhibit much stronger analogical reasoning at all three stages. Additionally, the strongest tested model, GPT-4V, performs better in tasks involving simple visual attributes like color and size, correlating with quicker human adult response times. Conversely, more complex tasks such as number, rotation, and reflection, which necessitate extensive cognitive processing and understanding of the 3D physical world, present more significant challenges. Altogether, these findings highlight the limitations of training models on data that primarily consists of 2D images and text.
Authors: Yifan Li, Yikai Wang, Yanwei Fu, Dongyu Ru, Zheng Zhang, Tong He
Abstract: Visual-Language Alignment (VLA) has gained a lot of attention since CLIP's groundbreaking work. Although CLIP performs well, the typical direct latent feature alignment lacks clarity in its representation and similarity scores. On the other hand, lexical representation, a vector whose element represents the similarity between the sample and a word from the vocabulary, is a natural sparse representation and interpretable, providing exact matches for individual words. However, lexical representations is difficult to learn due to no ground-truth supervision and false-discovery issues, and thus requires complex design to train effectively. In this paper, we introduce LexVLA, a more interpretable VLA framework by learning a unified lexical representation for both modalities without complex design. We use DINOv2 as our visual model for its local-inclined features and Llama 2, a generative language model, to leverage its in-context lexical prediction ability. To avoid the false discovery, we propose an overuse penalty to refrain the lexical representation from falsely frequently activating meaningless words. We demonstrate that these two pre-trained uni-modal models can be well-aligned by fine-tuning on modest multi-modal dataset and avoid intricate training configurations. On cross-modal retrieval benchmarks, LexVLA, trained on the CC-12M multi-modal dataset, outperforms baselines fine-tuned on larger datasets (e.g., YFCC15M) and those trained from scratch on even bigger datasets (e.g., 1.1B data, including CC-12M). We conduct extensive experiments to analyze LexVLA.
Authors: Lisanne van Gelderen, Cristian Tejedor-Garc\'ia
Abstract: Parkinson's disease (PD), the second most prevalent neurodegenerative disorder worldwide, frequently presents with early-stage speech impairments. Recent advancements in Artificial Intelligence (AI), particularly deep learning (DL), have significantly enhanced PD diagnosis through the analysis of speech data. Nevertheless, the progress of research is restricted by the limited availability of publicly accessible speech-based PD datasets, primarily due to privacy and ethical concerns. This review covers the latest DL-based AI approaches for speech-based PD classification, focusing on performance, available resources and associated challenges of 33 scientific works published between 2020 and March 2024. These DL approaches are categorized into end-to-end (E2E) learning, transfer learning (TL) and deep acoustic features (DAF) extraction. Among E2E approaches, Convolutional Neural Networks (CNNs) are prevalent, though Transformers are increasingly popular. E2E approaches face challenges such as limited data and computational resources, especially with Transformers. TL addresses these issues by providing more robust PD diagnosis and better generalizability across languages. DAF extraction aims to improve the explainability and interpretability of results by examining the specific effects of deep features on both other DL approaches and more traditional machine learning (ML) methods. However, it often underperforms compared to E2E and TL approaches. This review also discusses unresolved issues related to bias, explainability and privacy, highlighting the need for future research.
Authors: Wen Luo, Yu Xia, Shen Tianshu, Sujian Li
Abstract: The rise of social media and the exponential growth of multimodal communication necessitates advanced techniques for Multimodal Information Extraction (MIE). However, existing methodologies primarily rely on direct Image-Text interactions, a paradigm that often faces significant challenges due to semantic and modality gaps between images and text. In this paper, we introduce a new paradigm of Image-Context-Text interaction, where large multimodal models (LMMs) are utilized to generate descriptive textual context to bridge these gaps. In line with this paradigm, we propose a novel Shapley Value-based Contrastive Alignment (Shap-CA) method, which aligns both context-text and context-image pairs. Shap-CA initially applies the Shapley value concept from cooperative game theory to assess the individual contribution of each element in the set of contexts, texts and images towards total semantic and modality overlaps. Following this quantitative evaluation, a contrastive learning strategy is employed to enhance the interactive contribution within context-text/image pairs, while minimizing the influence across these pairs. Furthermore, we design an adaptive fusion module for selective cross-modal fusion. Extensive experiments across four MIE datasets demonstrate that our method significantly outperforms existing state-of-the-art methods.
Authors: Alex Kim, Maximilian Muhn, Valeri Nikolaev
Abstract: We investigate whether an LLM can successfully perform financial statement analysis in a way similar to a professional human analyst. We provide standardized and anonymous financial statements to GPT4 and instruct the model to analyze them to determine the direction of future earnings. Even without any narrative or industry-specific information, the LLM outperforms financial analysts in its ability to predict earnings changes. The LLM exhibits a relative advantage over human analysts in situations when the analysts tend to struggle. Furthermore, we find that the prediction accuracy of the LLM is on par with the performance of a narrowly trained state-of-the-art ML model. LLM prediction does not stem from its training memory. Instead, we find that the LLM generates useful narrative insights about a company's future performance. Lastly, our trading strategies based on GPT's predictions yield a higher Sharpe ratio and alphas than strategies based on other models. Taken together, our results suggest that LLMs may take a central role in decision-making.
Authors: Avanti Bhandarkar, Ronald Wilson, Anushka Swarup, Mengdi Zhu, Damon Woodard
Abstract: In the era of generative AI, the widespread adoption of Neural Text Generators (NTGs) presents new cybersecurity challenges, particularly within the realms of Digital Forensics and Incident Response (DFIR). These challenges primarily involve the detection and attribution of sources behind advanced attacks like spearphishing and disinformation campaigns. As NTGs evolve, the task of distinguishing between human and NTG-authored texts becomes critically complex. This paper rigorously evaluates the DFIR pipeline tailored for text-based security systems, specifically focusing on the challenges of detecting and attributing authorship of NTG-authored texts. By introducing a novel human-NTG co-authorship text attack, termed CS-ACT, our study uncovers significant vulnerabilities in traditional DFIR methodologies, highlighting discrepancies between ideal scenarios and real-world conditions. Utilizing 14 diverse datasets and 43 unique NTGs, up to the latest GPT-4, our research identifies substantial vulnerabilities in the forensic profiling phase, particularly in attributing authorship to NTGs. Our comprehensive evaluation points to factors such as model sophistication and the lack of distinctive style within NTGs as significant contributors for these vulnerabilities. Our findings underscore the necessity for more sophisticated and adaptable strategies, such as incorporating adversarial learning, stylizing NTGs, and implementing hierarchical attribution through the mapping of NTG lineages to enhance source attribution. This sets the stage for future research and the development of more resilient text-based security systems.
Authors: Jan Batzner, Volker Stocker, Stefan Schmid, Gjergji Kasneci
Abstract: LLMs are changing the way humans create and interact with content, potentially affecting citizens' political opinions and voting decisions. As LLMs increasingly shape our digital information ecosystems, auditing to evaluate biases, sycophancy, or steerability has emerged as an active field of research. In this paper, we evaluate and compare the alignment of six LLMs by OpenAI, Anthropic, and Cohere with German party positions and evaluate sycophancy based on a prompt experiment. We contribute to evaluating political bias and sycophancy in multi-party systems across major commercial LLMs. First, we develop the benchmark dataset GermanPartiesQA based on the Voting Advice Application Wahl-o-Mat covering 10 state and 1 national elections between 2021 and 2023. In our study, we find a left-green tendency across all examined LLMs. We then conduct our prompt experiment for which we use the benchmark and sociodemographic data of leading German parliamentarians to evaluate changes in LLMs responses. To differentiate between sycophancy and steerabilty, we use 'I am [politician X], ...' and 'You are [politician X], ...' prompts. Against our expectations, we do not observe notable differences between prompting 'I am' and 'You are'. While our findings underscore that LLM responses can be ideologically steered with political personas, they suggest that observed changes in LLM outputs could be better described as personalization to the given context rather than sycophancy.
Authors: Haoyu Chen, Wenbo Li, Jinjin Gu, Jingjing Ren, Sixiang Chen, Tian Ye, Renjing Pei, Kaiwen Zhou, Fenglong Song, Lei Zhu
Abstract: Natural images captured by mobile devices often suffer from multiple types of degradation, such as noise, blur, and low light. Traditional image restoration methods require manual selection of specific tasks, algorithms, and execution sequences, which is time-consuming and may yield suboptimal results. All-in-one models, though capable of handling multiple tasks, typically support only a limited range and often produce overly smooth, low-fidelity outcomes due to their broad data distribution fitting. To address these challenges, we first define a new pipeline for restoring images with multiple degradations, and then introduce RestoreAgent, an intelligent image restoration system leveraging multimodal large language models. RestoreAgent autonomously assesses the type and extent of degradation in input images and performs restoration through (1) determining the appropriate restoration tasks, (2) optimizing the task sequence, (3) selecting the most suitable models, and (4) executing the restoration. Experimental results demonstrate the superior performance of RestoreAgent in handling complex degradation, surpassing human experts. Furthermore, the system modular design facilitates the fast integration of new tasks and models, enhancing its flexibility and scalability for various applications.
Authors: Yannis Vasilakis, Rachel Bittner, Johan Pauwels
Abstract: Music two-tower multimodal systems integrate audio and text modalities into a joint audio-text space, enabling direct comparison between songs and their corresponding labels. These systems enable new approaches for classification and retrieval, leveraging both modalities. Despite the promising results they have shown for zero-shot classification and retrieval tasks, closer inspection of the embeddings is needed. This paper evaluates the inherent zero-shot properties of joint audio-text spaces for the case-study of instrument recognition. We present an evaluation and analysis of two-tower systems for zero-shot instrument recognition and a detailed analysis of the properties of the pre-joint and joint embeddings spaces. Our findings suggest that audio encoders alone demonstrate good quality, while challenges remain within the text encoder or joint space projection. Specifically, two-tower systems exhibit sensitivity towards specific words, favoring generic prompts over musically informed ones. Despite the large size of textual encoders, they do not yet leverage additional textual context or infer instruments accurately from their descriptions. Lastly, a novel approach for quantifying the semantic meaningfulness of the textual space leveraging an instrument ontology is proposed. This method reveals deficiencies in the systems' understanding of instruments and provides evidence of the need for fine-tuning text encoders on musical data.
Authors: Nikolhaus Howe, Micha{\l} Zajac, Ian McKenzie, Oskar Hollinsworth, Tom Tseng, Pierre-Luc Bacon, Adam Gleave
Abstract: Language model capabilities predictably improve from scaling a model's size and training data. Motivated by this, increasingly large language models have been trained, yielding an array of impressive capabilities. Yet these models are vulnerable to adversarial prompts, such as "jailbreaks" that hijack models to perform undesired behaviors, posing a significant risk of misuse. Prior work indicates that computer vision models become more robust with model and data scaling, raising the question: does language model robustness also improve with scale? We study this question empirically, finding that larger models respond substantially better to adversarial training, but there is little to no benefit from model scale in the absence of explicit defenses.
Authors: Yuxiao Qu, Tianjun Zhang, Naman Garg, Aviral Kumar
Abstract: A central piece in enabling intelligent agentic behavior in foundation models is to make them capable of introspecting upon their behavior, reasoning, and correcting their mistakes as more computation or interaction is available. Even the strongest proprietary large language models (LLMs) do not quite exhibit the ability of continually improving their responses sequentially, even in scenarios where they are explicitly told that they are making a mistake. In this paper, we develop RISE: Recursive IntroSpEction, an approach for fine-tuning LLMs to introduce this capability, despite prior work hypothesizing that this capability may not be possible to attain. Our approach prescribes an iterative fine-tuning procedure, which attempts to teach the model how to alter its response after having executed previously unsuccessful attempts to solve a hard test-time problem, with optionally additional environment feedback. RISE poses fine-tuning for a single-turn prompt as solving a multi-turn Markov decision process (MDP), where the initial state is the prompt. Inspired by principles in online imitation learning and reinforcement learning, we propose strategies for multi-turn data collection and training so as to imbue an LLM with the capability to recursively detect and correct its previous mistakes in subsequent iterations. Our experiments show that RISE enables Llama2, Llama3, and Mistral models to improve themselves with more turns on math reasoning tasks, outperforming several single-turn strategies given an equal amount of inference-time computation. We also find that RISE scales well, often attaining larger benefits with more capable models. Our analysis shows that RISE makes meaningful improvements to responses to arrive at the correct solution for challenging prompts, without disrupting one-turn abilities as a result of expressing more complex distributions.
Authors: Zhengbo Wang, Jian Liang
Abstract: Low-Rank Adaptation, also known as LoRA, has emerged as a prominent method for parameter-efficient fine-tuning foundation models by re-parameterizing the original matrix into the product of two low-rank matrices. Despite its efficiency, LoRA often yields inferior performance compared to full fine-tuning. In this paper, we propose LoRA-Pro to bridge this performance gap. Firstly, we delve into the optimization processes in LoRA and full fine-tuning. We reveal that while LoRA employs low-rank approximation, it neglects to approximate the optimization process of full fine-tuning. To address this, we introduce a novel concept called the "equivalent gradient." This virtual gradient makes the optimization process on the re-parameterized matrix equivalent to LoRA, which can be used to quantify the differences between LoRA and full fine-tuning. The equivalent gradient is derived from the gradients of matrices $A$ and $B$. To narrow the performance gap, our approach minimizes the differences between the equivalent gradient and the gradient obtained from full fine-tuning during the optimization process. By solving this objective, we derive optimal closed-form solutions for updating matrices $A$ and $B$. Our method constrains the optimization process, shrinking the performance gap between LoRA and full fine-tuning. Extensive experiments on natural language processing tasks validate the effectiveness of our method.
Authors: Hongseok Choi, Hyunju Lee
Abstract: To improve deep-learning performance in low-resource settings, many researchers have redesigned model architectures or applied additional data (e.g., external resources, unlabeled samples). However, there have been relatively few discussions on how to make good use of small amounts of labeled samples, although it is potentially beneficial and should be done before applying additional data or redesigning models. In this study, we assume a low-resource setting in which only a few labeled samples (i.e., 30-100 per class) are available, and we discuss how to exploit them without additional data or model redesigns. We explore possible approaches in the following three aspects: training-validation splitting, early stopping, and weight initialization. Extensive experiments are conducted on six public sentence classification datasets. Performance on various evaluation metrics (e.g., accuracy, loss, and calibration error) significantly varied depending on the approaches that were combined in the three aspects. Based on the results, we propose an integrated method, which is to initialize the model with a weight averaging method and use a non-validation stop method to train all samples. This simple integrated method consistently outperforms the competitive methods; e.g., the average accuracy of six datasets of this method was 1.8% higher than those of conventional validation-based methods. In addition, the integrated method further improves the performance when adapted to several state-of-the-art models that use additional data or redesign the network architecture (e.g., self-training and enhanced structural models). Our results highlight the importance of the training strategy and suggest that the integrated method can be the first step in the low-resource setting. This study provides empirical knowledge that will be helpful when dealing with low-resource data in future efforts.
Authors: Qixiang Fang, Anastasia Giachanou, Ayoub Bagheri
Abstract: Stance detection (SD) concerns automatically determining the viewpoint (i.e., in favour of, against, or neutral) of a text's author towards a target. SD has been applied to many research topics, among which the detection of stances behind political tweets is an important one. In this paper, we apply SD to a dataset of tweets from official party accounts in the Netherlands between 2017 and 2021, with a focus on stances towards traditional gender role division, a dividing issue between (some) Dutch political parties. To implement and improve SD of traditional gender role division, we propose to leverage an established survey instrument from social sciences, which has been validated for the purpose of measuring attitudes towards traditional gender role division. Based on our experiments, we show that using such a validated survey instrument helps to improve SD performance.
Authors: Joe Stacey, Marek Rei
Abstract: Knowledge distillation optimises a smaller student model to behave similarly to a larger teacher model, retaining some of the performance benefits. While this method can improve results on in-distribution examples, it does not necessarily generalise to out-of-distribution (OOD) settings. We investigate two complementary methods for improving the robustness of the resulting student models on OOD domains. The first approach augments the distillation with generated unlabelled examples that match the target distribution. The second method upsamples data points among the training set that are similar to the target distribution. When applied on the task of natural language inference (NLI), our experiments on MNLI show that distillation with these modifications outperforms previous robustness solutions. We also find that these methods improve performance on OOD domains even beyond the target domain.
Authors: Le Zhuo, Ruibin Yuan, Jiahao Pan, Yinghao Ma, Yizhi LI, Ge Zhang, Si Liu, Roger Dannenberg, Jie Fu, Chenghua Lin, Emmanouil Benetos, Wei Xue, Yike Guo
Abstract: We introduce LyricWhiz, a robust, multilingual, and zero-shot automatic lyrics transcription method achieving state-of-the-art performance on various lyrics transcription datasets, even in challenging genres such as rock and metal. Our novel, training-free approach utilizes Whisper, a weakly supervised robust speech recognition model, and GPT-4, today's most performant chat-based large language model. In the proposed method, Whisper functions as the "ear" by transcribing the audio, while GPT-4 serves as the "brain," acting as an annotator with a strong performance for contextualized output selection and correction. Our experiments show that LyricWhiz significantly reduces Word Error Rate compared to existing methods in English and can effectively transcribe lyrics across multiple languages. Furthermore, we use LyricWhiz to create the first publicly available, large-scale, multilingual lyrics transcription dataset with a CC-BY-NC-SA copyright license, based on MTG-Jamendo, and offer a human-annotated subset for noise level estimation and evaluation. We anticipate that our proposed method and dataset will advance the development of multilingual lyrics transcription, a challenging and emerging task.
Authors: Lucas Bandarkar, Davis Liang, Benjamin Muller, Mikel Artetxe, Satya Narayan Shukla, Donald Husa, Naman Goyal, Abhinandan Krishnan, Luke Zettlemoyer, Madian Khabsa
Abstract: We present Belebele, a multiple-choice machine reading comprehension (MRC) dataset spanning 122 language variants. Significantly expanding the language coverage of natural language understanding (NLU) benchmarks, this dataset enables the evaluation of text models in high-, medium-, and low-resource languages. Each question is based on a short passage from the Flores-200 dataset and has four multiple-choice answers. The questions were carefully curated to discriminate between models with different levels of general language comprehension. The English dataset on its own proves difficult enough to challenge state-of-the-art language models. Being fully parallel, this dataset enables direct comparison of model performance across all languages. We use this dataset to evaluate the capabilities of multilingual masked language models (MLMs) and large language models (LLMs). We present extensive results and find that despite significant cross-lingual transfer in English-centric LLMs, much smaller MLMs pretrained on balanced multilingual data still understand far more languages. We also observe that larger vocabulary size and conscious vocabulary construction correlate with better performance on low-resource languages. Overall, Belebele opens up new avenues for evaluating and analyzing the multilingual capabilities of NLP systems.
Authors: Daixuan Cheng, Shaohan Huang, Furu Wei
Abstract: We explore how continued pre-training on domain-specific corpora influences large language models, revealing that training on the raw corpora endows the model with domain knowledge, but drastically hurts its prompting ability for question answering. Taken inspiration from human learning via reading comprehension--practice after reading improves the ability to answer questions based on the learned knowledge--we propose a simple method for transforming raw corpora into reading comprehension texts. Each raw text is enriched with a series of tasks related to its content. Our method, highly scalable and applicable to any pre-training corpora, consistently enhances performance across various tasks in three different domains: biomedicine, finance, and law. Notably, our 7B language model achieves competitive performance with domain-specific models of much larger scales, such as BloombergGPT-50B. Furthermore, we demonstrate that domain-specific reading comprehension texts can improve the model's performance even on general benchmarks, showing the potential to develop a general model across even more domains. Our model, code, and data are available at https://github.com/microsoft/LMOps.
Authors: Anh Thi-Hoang Nguyen, Dung Ha Nguyen, Nguyet Thi Nguyen, Khanh Thanh-Duy Ho, Kiet Van Nguyen
Abstract: Social media data is a valuable resource for research, yet it contains a wide range of non-standard words (NSW). These irregularities hinder the effective operation of NLP tools. Current state-of-the-art methods for the Vietnamese language address this issue as a problem of lexical normalization, involving the creation of manual rules or the implementation of multi-staged deep learning frameworks, which necessitate extensive efforts to craft intricate rules. In contrast, our approach is straightforward, employing solely a sequence-to-sequence (Seq2Seq) model. In this research, we provide a dataset for textual normalization, comprising 2,181 human-annotated comments with an inter-annotator agreement of 0.9014. By leveraging the Seq2Seq model for textual normalization, our results reveal that the accuracy achieved falls slightly short of 70%. Nevertheless, textual normalization enhances the accuracy of the Hate Speech Detection (HSD) task by approximately 2%, demonstrating its potential to improve the performance of complex NLP tasks. Our dataset is accessible for research purposes.
Authors: Chen Zhang, Dawei Song, Zheyu Ye, Yan Gao
Abstract: Language model (LM) distillation is a trending area that aims to distil the knowledge residing in a large teacher LM to a small student one. While various methods have been proposed to maximize the effectiveness of the distillation, significant challenges persist, particularly when there is a substantial capacity gap between the teacher and student LMs. This issue, often referred to as the \textit{curse} of capacity gap, suggests that a larger teacher does not necessarily result in a superior student compared to one distilled from a smaller teacher. In other words, there is likely an optimal teacher yielding the best student along the scaling course of the teacher. However, the curse of capacity gap can not be tackled without notable compute overhead, as indicated in previous studies. In the context of large LMs (LLMs), previously viable approaches become much less meaningful, as it is an impossible triangle to distill an expected student from an optimal teacher student with small compute overhead. Fortunately, the impossible triangle can fortunately be possible provided an inducted \textit{law} of capacity gap. In this paper, we take the spirits of scaling law and reveal that the optimal teacher scale almost consistently follows a linear scaling with the student scale across different model architectures and data scales. The law later guides us to distil a 3B student LM (termed \textsc{MiniMA}) from LLaMA2-7B. \textsc{MiniMA} is demonstrated to outperform a wide range of 3B competitors and could even compete with several 7B models.
Authors: Nicolae-Catalin Ristea, Andrei Anghel, Radu Tudor Ionescu
Abstract: Speech classification tasks often require powerful language understanding models to grasp useful features, which becomes problematic when limited training data is available. To attain superior classification performance, we propose to harness the inherent value of multimodal representations by transcribing speech using automatic speech recognition (ASR) models and translating the transcripts into different languages via pretrained translation models. We thus obtain an audio-textual (multimodal) representation for each data sample. Subsequently, we combine language-specific Bidirectional Encoder Representations from Transformers (BERT) with Wav2Vec2.0 audio features via a novel cascaded cross-modal transformer (CCMT). Our model is based on two cascaded transformer blocks. The first one combines text-specific features from distinct languages, while the second one combines acoustic features with multilingual features previously learned by the first transformer block. We employed our system in the Requests Sub-Challenge of the ACM Multimedia 2023 Computational Paralinguistics Challenge. CCMT was declared the winning solution, obtaining an unweighted average recall (UAR) of 65.41% and 85.87% for complaint and request detection, respectively. Moreover, we applied our framework on the Speech Commands v2 and HarperValleyBank dialog data sets, surpassing previous studies reporting results on these benchmarks. Our code is freely available for download at: https://github.com/ristea/ccmt.
Authors: Zirui Liu, Jiayi Yuan, Hongye Jin, Shaochen Zhong, Zhaozhuo Xu, Vladimir Braverman, Beidi Chen, Xia Hu
Abstract: Efficiently serving large language models (LLMs) requires batching of many requests to reduce the cost per request. Yet, with larger batch sizes and longer context lengths, the key-value (KV) cache, which stores attention keys and values to avoid re-computations, significantly increases memory demands and becomes the new bottleneck in speed and memory usage. Additionally, the loading of the KV cache causes the computational core to be idle, which limits the inference speed. A straightforward and effective solution to reduce KV cache size is quantization, which decreases the total bytes taken by KV cache. However, there is a lack of in-depth studies that explore the element distribution of KV cache to understand the hardness and limitation of KV cache quantization. To fill the gap, we conducted a comprehensive study on the element distribution in KV cache of popular LLMs. Our findings indicate that the key cache should be quantized per-channel, i.e., group elements along the channel dimension and quantize them together. In contrast, the value cache should be quantized per-token. From this analysis, we developed a tuning-free 2bit KV cache quantization algorithm named KIVI. With hardware-friendly implementation, KIVI can enable Llama, Falcon, and Mistral models to maintain almost the same quality while using $\mathbf{2.6\times}$ less peak memory (including model weight). This reduction in memory usage enables up to $\mathbf{4\times}$ larger batch size, bringing $\mathbf{2.35\times \sim 3.47\times}$ throughput on real LLM inference workload. The source code is available at https://github.com/jy-yuan/KIVI.
Authors: Qingkai Zeng, Yuyang Bai, Zhaoxuan Tan, Shangbin Feng, Zhenwen Liang, Zhihan Zhang, Meng Jiang
Abstract: Automatic taxonomy induction is crucial for web search, recommendation systems, and question answering. Manual curation of taxonomies is expensive in terms of human effort, making automatic taxonomy construction highly desirable. In this work, we introduce Chain-of-Layer which is an in-context learning framework designed to induct taxonomies from a given set of entities. Chain-of-Layer breaks down the task into selecting relevant candidate entities in each layer and gradually building the taxonomy from top to bottom. To minimize errors, we introduce the Ensemble-based Ranking Filter to reduce the hallucinated content generated at each iteration. Through extensive experiments, we demonstrate that Chain-of-Layer achieves state-of-the-art performance on four real-world benchmarks.
Authors: Jie Ren, Qipeng Guo, Hang Yan, Dongrui Liu, Quanshi Zhang, Xipeng Qiu, Dahua Lin
Abstract: Although large language models (LLMs) have demonstrated remarkable performance, the lack of transparency in their inference logic raises concerns about their trustworthiness. To gain a better understanding of LLMs, we conduct a detailed analysis of the operations of attention heads and aim to better understand the in-context learning of LLMs. Specifically, we investigate whether attention heads encode two types of relationships between tokens present in natural languages: the syntactic dependency parsed from sentences and the relation within knowledge graphs. We find that certain attention heads exhibit a pattern where, when attending to head tokens, they recall tail tokens and increase the output logits of those tail tokens. More crucially, the formulation of such semantic induction heads has a close correlation with the emergence of the in-context learning ability of language models. The study of semantic attention heads advances our understanding of the intricate operations of attention heads in transformers, and further provides new insights into the in-context learning of LLMs.
Authors: Lilong Xue, Dan Zhang, Yuxiao Dong, Jie Tang
Abstract: Large Language Models (LLMs) have demonstrated exceptional abilities in comprehending and generating text, motivating numerous researchers to utilize them for Information Extraction (IE) purposes, including Relation Extraction (RE). Nonetheless, most existing methods are predominantly designed for Sentence-level Relation Extraction (SentRE) tasks, which typically encompass a restricted set of relations and triplet facts within a single sentence. Furthermore, certain approaches resort to treating relations as candidate choices integrated into prompt templates, leading to inefficient processing and suboptimal performance when tackling Document-Level Relation Extraction (DocRE) tasks, which entail handling multiple relations and triplet facts distributed across a given document, posing distinct challenges. To overcome these limitations, we introduce AutoRE, an end-to-end DocRE model that adopts a novel RE extraction paradigm named RHF (Relation-Head-Facts). Unlike existing approaches, AutoRE does not rely on the assumption of known relation options, making it more reflective of real-world scenarios. Additionally, we have developed an easily extensible RE framework using a Parameters Efficient Fine Tuning (PEFT) algorithm (QLoRA). Our experiments on the RE-DocRED dataset showcase AutoRE's best performance, achieving state-of-the-art results, surpassing TAG by 10.03\% and 9.03\% respectively on the dev and test set. The code is available\url{https://github.com/THUDM/AutoRE} and the demonstration video is provided https://www.youtube.com/watch?v=IhKRsZUAxKk
URLs: https://github.com/THUDM/AutoRE, https://www.youtube.com/watch?v=IhKRsZUAxKk
Authors: Qixiang Fang, Daniel L. Oberski, Dong Nguyen
Abstract: Many existing benchmarks of large (multimodal) language models (LLMs) focus on measuring LLMs' academic proficiency, often with also an interest in comparing model performance with human test takers. While these benchmarks have proven key to the development of LLMs, they suffer from several limitations, including questionable measurement quality (e.g., Do they measure what they are supposed to in a reliable way?), lack of quality assessment on the item level (e.g., Are some items more important or difficult than others?) and unclear human population reference (e.g., To whom can the model be compared?). In response to these challenges, we propose leveraging knowledge from psychometrics - a field dedicated to the measurement of latent variables like academic proficiency - into LLM benchmarking. We make three primary contributions. First, we introduce PATCH: a novel framework for {P}sychometrics-{A}ssis{T}ed ben{CH}marking of LLMs. PATCH addresses the aforementioned limitations, presenting a new direction for LLM benchmark research. Second, we implement PATCH by measuring GPT-4 and Gemini-Pro-Vision's proficiency in 8th grade mathematics against 56 human populations. We show that adopting a psychometrics-based approach yields evaluation outcomes that diverge from those based on existing benchmarking practices. Third, we release 4 high-quality datasets to support measuring and comparing LLM proficiency in grade school mathematics and science against human populations.
Authors: Hye Sun Yun, David Pogrebitskiy, Iain J. Marshall, Byron C. Wallace
Abstract: Meta-analyses statistically aggregate the findings of different randomized controlled trials (RCTs) to assess treatment effectiveness. Because this yields robust estimates of treatment effectiveness, results from meta-analyses are considered the strongest form of evidence. However, rigorous evidence syntheses are time-consuming and labor-intensive, requiring manual extraction of data from individual trials to be synthesized. Ideally, language technologies would permit fully automatic meta-analysis, on demand. This requires accurately extracting numerical results from individual trials, which has been beyond the capabilities of natural language processing (NLP) models to date. In this work, we evaluate whether modern large language models (LLMs) can reliably perform this task. We annotate (and release) a modest but granular evaluation dataset of clinical trial reports with numerical findings attached to interventions, comparators, and outcomes. Using this dataset, we evaluate the performance of seven LLMs applied zero-shot for the task of conditionally extracting numerical findings from trial reports. We find that massive LLMs that can accommodate lengthy inputs are tantalizingly close to realizing fully automatic meta-analysis, especially for dichotomous (binary) outcomes (e.g., mortality). However, LLMs -- including ones trained on biomedical texts -- perform poorly when the outcome measures are complex and tallying the results requires inference. This work charts a path toward fully automatic meta-analysis of RCTs via LLMs, while also highlighting the limitations of existing models for this aim.
Authors: Haoran You, Yichao Fu, Zheng Wang, Amir Yazdanbakhsh, Yingyan Celine Lin
Abstract: Autoregressive Large Language Models (LLMs) have achieved impressive performance in language tasks but face two significant bottlenecks: (1) quadratic complexity in the attention module as the number of tokens increases, and (2) limited efficiency due to the sequential processing nature of autoregressive LLMs during generation. While linear attention and speculative decoding offer potential solutions, their applicability and synergistic potential for enhancing autoregressive LLMs remain uncertain. We conduct the first comprehensive study on the efficacy of existing linear attention methods for autoregressive LLMs, integrating them with speculative decoding. We introduce an augmentation technique for linear attention that ensures compatibility with speculative decoding, enabling more efficient training and serving of LLMs. Extensive experiments and ablation studies involving seven existing linear attention models and five encoder/decoder-based LLMs consistently validate the effectiveness of our augmented linearized LLMs. Notably, our approach achieves up to a 6.67 reduction in perplexity on the LLaMA model and up to a 2$\times$ speedup during generation compared to prior linear attention methods. Codes and models are available at https://github.com/GATECH-EIC/Linearized-LLM.
Authors: Aditya Kalyanpur, Kailash Karthik Saravanakumar, Victor Barres, CJ McFate, Lori Moon, Nati Seifu, Maksim Eremeev, Jose Barrera, Abraham Bautista-Castillo, Eric Brown, David Ferrucci
Abstract: The advent of Large Language Models (LLMs) and Generative AI has revolutionized natural language applications across various domains. However, high-stakes decision-making tasks in fields such as medical, legal and finance require a level of precision, comprehensiveness, and logical consistency that pure LLM or Retrieval-Augmented-Generation (RAG) approaches often fail to deliver. At Elemental Cognition (EC), we have developed a neuro-symbolic AI platform to tackle these problems. The platform integrates fine-tuned LLMs for knowledge extraction and alignment with a robust symbolic reasoning engine for logical inference, planning and interactive constraint solving. We describe Cora, a Collaborative Research Assistant built on this platform, that is designed to perform complex research and discovery tasks in high-stakes domains. This paper discusses the multi-step inference challenges inherent in such domains, critiques the limitations of existing LLM-based methods, and demonstrates how Cora's neuro-symbolic approach effectively addresses these issues. We provide an overview of the system architecture, key algorithms for knowledge extraction and formal reasoning, and present preliminary evaluation results that highlight Cora's superior performance compared to well-known LLM and RAG baselines.
Authors: Haibo Jin, Leyang Hu, Xinuo Li, Peiyan Zhang, Chonghan Chen, Jun Zhuang, Haohan Wang
Abstract: The rapid evolution of artificial intelligence (AI) through developments in Large Language Models (LLMs) and Vision-Language Models (VLMs) has brought significant advancements across various technological domains. While these models enhance capabilities in natural language processing and visual interactive tasks, their growing adoption raises critical concerns regarding security and ethical alignment. This survey provides an extensive review of the emerging field of jailbreaking--deliberately circumventing the ethical and operational boundaries of LLMs and VLMs--and the consequent development of defense mechanisms. Our study categorizes jailbreaks into seven distinct types and elaborates on defense strategies that address these vulnerabilities. Through this comprehensive examination, we identify research gaps and propose directions for future studies to enhance the security frameworks of LLMs and VLMs. Our findings underscore the necessity for a unified perspective that integrates both jailbreak strategies and defensive solutions to foster a robust, secure, and reliable environment for the next generation of language models. More details can be found on our website: \url{https://chonghan-chen.com/llm-jailbreak-zoo-survey/}.
Authors: Weiming Li, Manni Duan, Dong An, Yan Shao
Abstract: Large language models (LLMs) demonstrate extraordinary abilities in a wide range of natural language processing (NLP) tasks. In this paper, we show that, beyond text understanding capability, LLMs are capable of processing text layouts that are denoted by spatial markers. They are able to answer questions that require explicit spatial perceiving and reasoning, while a drastic performance drop is observed when the spatial markers from the original data are excluded. We perform a series of experiments with the GPT-3.5, Baichuan2, Llama2 and ChatGLM3 models on various types of layout-sensitive datasets for further analysis. The experimental results reveal that the layout understanding ability of LLMs is mainly introduced by the coding data for pretraining, which is further enhanced at the instruction-tuning stage. In addition, layout understanding can be enhanced by integrating low-cost, auto-generated data approached by a novel text game. Finally, we show that layout understanding ability is beneficial for building efficient visual question-answering (VQA) systems.
Authors: Ping Yu, Jing Xu, Jason Weston, Ilia Kulikov
Abstract: Large language models (LLMs) can spend extra compute during inference to generate intermediate thoughts, which helps to produce better final responses. Since Chain-of-Thought (Wei et al., 2022), many such System 2 techniques have been proposed such as Rephrase and Respond (Deng et al., 2023a), System 2 Attention (Weston and Sukhbaatar, 2023) and Branch-Solve-Merge (Saha et al., 2023). In this work we investigate self-supervised methods to ``compile'' (distill) higher quality outputs from System 2 techniques back into LLM generations without intermediate reasoning token sequences, as this reasoning has been distilled into System 1. We show that several such techniques can be successfully distilled, resulting in improved results compared to the original System 1 performance, and with less inference cost than System 2. We posit that such System 2 distillation will be an important feature of future continually learning AI systems, enabling them to focus System 2 capabilities on the reasoning tasks that they cannot yet do well.
Authors: Songyang Zhang, Chuyu Zhang, Yingfan Hu, Haowen Shen, Kuikun Liu, Zerun Ma, Fengzhe Zhou, Wenwei Zhang, Xuming He, Dahua Lin, Kai Chen
Abstract: While LLM-Based agents, which use external tools to solve complex problems, have made significant progress, benchmarking their ability is challenging, thereby hindering a clear understanding of their limitations. In this paper, we propose an interactive evaluation framework, named CIBench, to comprehensively assess LLMs' ability to utilize code interpreters for data science tasks. Our evaluation framework includes an evaluation dataset and two evaluation modes. The evaluation dataset is constructed using an LLM-human cooperative approach and simulates an authentic workflow by leveraging consecutive and interactive IPython sessions. The two evaluation modes assess LLMs' ability with and without human assistance. We conduct extensive experiments to analyze the ability of 24 LLMs on CIBench and provide valuable insights for future LLMs in code interpreter utilization.
Authors: Shaomang Huang, Jianfeng Pan, Hanzhong Zheng
Abstract: In the domain of Large Language Model (LLM), LLMs demonstrate significant capabilities in natural language understanding and generation. With the growing needs of applying LLMs on various domains, it is a research question that how to efficiently train and build a model that has expertise in different domains but with a low training cost. We propose CCoE architecture, a framework of easily coupling multiple strong domain experts together to fuse into a big LLM, provides a collective way of utilizing the different domain expert LLMs. Besides, training a large collaborative of multiple expert LLMs requires a high requirements on training sources. CCoE bypasses this problem through isolating other experts and train each expert separately. The design of CCoE assembles multiple expert LLMs through the CoE (Collaboration of Experts) layer. Each CoE layer could have one or more expert LLMs. Expert LLMs have different number of layers and have been well-trained for different domain tasks. Each expert is fine-tuned to be able to achieve the comparable results with SOTA domain LLMs. We start from 5 experts in the domain of Code, Math, Law, text-to-SQL and Medical. The results indicate that our CCoE framework can easily and efficiently boost nearly 10%-20% performance on original base model in different domains but using less resources on training, as well as inference.
Authors: Jinghui Zhang, Dandan Qiao, Mochen Yang, Qiang Wei
Abstract: What happens if we train a new Large Language Model (LLM) using data that are at least partially generated by other LLMs? The explosive success of LLMs means that a substantial amount of content online will be generated by LLMs rather than humans, which will inevitably enter the training datasets of next-generation LLMs. We evaluate the implications of such "regurgitative training" on LLM performance. Through fine-tuning GPT-3.5 with data generated either by itself or by other LLMs in a machine translation task, we find strong evidence that regurgitative training clearly handicaps the performance of LLMs. The same performance loss of regurgitative training is observed on transformer models that we train from scratch. We find suggestive evidence that the performance disadvantage of regurgitative training can be attributed to at least two mechanisms: (1) higher error rates and (2) lower lexical diversity in LLM-generated data as compared to real data. Based on these mechanisms, we propose and evaluate three different strategies to mitigate the performance loss of regurgitative training. First, we devise data-driven metrics to gauge the quality of each LLM-generated data instance, and then carry out an ordered training process where high-quality data are added before low-quality ones. Second, we combine data generated by multiple different LLMs (as an attempt to increase lexical diversity). Third, we train an AI detection classifier to differentiate between LLM- and human-generated data, and include LLM-generated data in the order of resemblance to human-generated data. All three strategies can improve the performance of regurgitative training to some extent but are not always able to fully close the gap from training with real data. Our results highlight the value of real, human-generated data in training LLMs, which cannot be easily substituted by synthetic, LLM-generated data.
Authors: Danni Yu, Marina Bondi, Ken Hyland
Abstract: One of the most powerful and enduring ideas in written discourse analysis is that genres can be described in terms of the moves which structure a writer's purpose. Considerable research has sought to identify these distinct communicative acts, but analyses have been beset by problems of subjectivity, reliability and the time-consuming need for multiple coders to confirm analyses. In this paper we employ the affordances of GPT-4 to automate the annotation process by using natural language prompts. Focusing on abstracts from articles in four applied linguistics journals, we devise prompts which enable the model to identify moves effectively. The annotated outputs of these prompts were evaluated by two assessors with a third addressing disagreements. The results show that an 8-shot prompt was more effective than one using two, confirming that the inclusion of examples illustrating areas of variability can enhance GPT-4's ability to recognize multiple moves in a single sentence and reduce bias related to textual position. We suggest that GPT-4 offers considerable potential in automating this annotation process, when human actors with domain specific linguistic expertise inform the prompting process.
Authors: Kenza Benkirane, Laura Gongas, Shahar Pelles, Naomi Fuchs, Joshua Darmon, Pontus Stenetorp, David Ifeoluwa Adelani, Eduardo S\'anchez
Abstract: Recent advancements in massively multilingual machine translation systems have significantly enhanced translation accuracy; however, even the best performing systems still generate hallucinations, severely impacting user trust. Detecting hallucinations in Machine Translation (MT) remains a critical challenge, particularly since existing methods excel with High-Resource Languages (HRLs) but exhibit substantial limitations when applied to Low-Resource Languages (LRLs). This paper evaluates hallucination detection approaches using Large Language Models (LLMs) and semantic similarity within massively multilingual embeddings. Our study spans 16 language directions, covering HRLs, LRLs, with diverse scripts. We find that the choice of model is essential for performance. On average, for HRLs, Llama3-70B outperforms the previous state of the art by as much as 0.16 MCC (Matthews Correlation Coefficient). However, for LRLs we observe that Claude Sonnet outperforms other LLMs on average by 0.03 MCC. The key takeaway from our study is that LLMs can achieve performance comparable or even better than previously proposed models, despite not being explicitly trained for any machine translation task. However, their advantage is less significant for LRLs.
Authors: Jonathan Hayase, Alisa Liu, Yejin Choi, Sewoong Oh, Noah A. Smith
Abstract: The pretraining data of today's strongest language models is opaque; in particular, little is known about the proportions of various domains or languages represented. In this work, we tackle a task which we call data mixture inference, which aims to uncover the distributional make-up of training data. We introduce a novel attack based on a previously overlooked source of information -- byte-pair encoding (BPE) tokenizers, used by the vast majority of modern language models. Our key insight is that the ordered list of merge rules learned by a BPE tokenizer naturally reveals information about the token frequencies in its training data: the first merge is the most common byte pair, the second is the most common pair after merging the first token, and so on. Given a tokenizer's merge list along with data samples for each category of interest, we formulate a linear program that solves for the proportion of each category in the tokenizer's training set. Importantly, to the extent to which tokenizer training data is representative of the pretraining data, we indirectly learn about pretraining data. In controlled experiments, we show that our attack recovers mixture ratios with high precision for tokenizers trained on known mixtures of natural languages, programming languages, and data sources. We then apply our approach to off-the-shelf tokenizers released with recent LMs. We confirm much publicly disclosed information about these models, and also make several new inferences: GPT-4o's tokenizer is much more multilingual than its predecessors, training on 39% non-English data; Llama3 extends GPT-3.5's tokenizer primarily for multilingual (48%) use; GPT-3.5's and Claude's tokenizers are trained on predominantly code (~60%). We hope our work sheds light on current design practices for pretraining data, and inspires continued research into data mixture inference for LMs.
Authors: Yixiu Liu, Yuxiang Zheng, Shijie Xia, Yuan Guo, Jiajun Li, Yi Tu, Chaoling Song, Pengfei Liu
Abstract: The deployment of Large Language Models (LLMs) in content generation raises significant safety concerns, particularly regarding the transparency and interpretability of content evaluations. Current methods, primarily focused on binary safety classifications, lack mechanisms for detailed critique, limiting their utility for model improvement and user trust. To address these limitations, we introduce SAFETY-J, a bilingual generative safety evaluator for English and Chinese with critique-based judgment. SAFETY-J utilizes a robust training dataset that includes diverse dialogues and augmented query-response pairs to assess safety across various scenarios comprehensively. We establish an automated meta-evaluation benchmark that objectively assesses the quality of critiques with minimal human intervention, facilitating scalable and continuous improvement. Additionally, SAFETY-J employs an iterative preference learning technique to dynamically refine safety assessments based on meta-evaluations and critiques. Our evaluations demonstrate that SAFETY-J provides more nuanced and accurate safety evaluations, thereby enhancing both critique quality and predictive reliability in complex content scenarios. To facilitate further research and application, we open-source SAFETY-J's training protocols, datasets, and code at \url{https://github.com/GAIR-NLP/Safety-J}.
Authors: Anastasiia Sedova, Robert Litschko, Diego Frassinelli, Benjamin Roth, Barbara Plank
Abstract: One of the major aspects contributing to the striking performance of large language models (LLMs) is the vast amount of factual knowledge accumulated during pre-training. Yet, many LLMs suffer from self-inconsistency, which raises doubts about their trustworthiness and reliability. In this paper, we focus on entity type ambiguity and analyze current state-of-the-art LLMs for their proficiency and consistency in applying their factual knowledge when prompted for entities under ambiguity. To do so, we propose an evaluation protocol that disentangles knowing from applying knowledge, and test state-of-the-art LLMs on 49 entities. Our experiments reveal that LLMs perform poorly with ambiguous prompts, achieving only 80% accuracy. Our results further demonstrate systematic discrepancies in LLM behavior and their failure to consistently apply information, indicating that the models can exhibit knowledge without being able to utilize it, significant biases for preferred readings, as well as self inconsistencies. Our study highlights the importance of handling entity ambiguity in future for more trustworthy LLMs
Authors: Ziqing Ji, Pranav Kulkarni, Marko Neskovic, Kevin Nolan, Yan Xu
Abstract: With news and information being as easy to access as they currently are, it is more important than ever to ensure that people are not mislead by what they read. Recently, the rise of neural fake news (AI-generated fake news) and its demonstrated effectiveness at fooling humans has prompted the development of models to detect it. One such model is the Grover model, which can both detect neural fake news to prevent it, and generate it to demonstrate how a model could be misused to fool human readers. In this work we explore the Grover model's fake news detection capabilities by performing targeted attacks through perturbations on input news articles. Through this we test Grover's resilience to these adversarial attacks and expose some potential vulnerabilities which should be addressed in further iterations to ensure it can detect all types of fake news accurately.
Authors: J. Cancellieri, W. Didimo, A. Fronzetti Colladon, F. Montecchiani, R. Vestrelli
Abstract: This paper presents a new decision support system offered for an in-depth analysis of semantic networks, which can provide insights for a better exploration of a brand's image and the improvement of its connectivity. In terms of network analysis, we show that this goal is achieved by solving an extended version of the Maximum Betweenness Improvement problem, which includes the possibility of considering adversarial nodes, constrained budgets, and weighted networks - where connectivity improvement can be obtained by adding links or increasing the weight of existing connections. Our contribution includes a new algorithmic framework and the integration of this framework into a software system called Brand Network Booster (BNB), which supports brand connectivity evaluation and improvement. We present this new system together with three case studies, and we also discuss its performance. Our tool and approach are valuable to both network scholars and in facilitating strategic decision-making processes for marketing and communication managers across various sectors, be it public or private.
Authors: Shengyao Zhuang, Bevan Koopman, Xiaoran Chu, Guido Zuccon
Abstract: The emergence of Vec2Text -- a method for text embedding inversion -- has raised serious privacy concerns for dense retrieval systems which use text embeddings, such as those offered by OpenAI and Cohere. This threat comes from the ability for a malicious attacker with access to embeddings to reconstruct the original text. In this paper, we investigate various factors related to embedding models that may impact text recoverability via Vec2Text. We explore factors such as distance metrics, pooling functions, bottleneck pre-training, training with noise addition, embedding quantization, and embedding dimensions, which were not considered in the original Vec2Text paper. Through a comprehensive analysis of these factors, our objective is to gain a deeper understanding of the key elements that affect the trade-offs between the text recoverability and retrieval effectiveness of dense retrieval systems, offering insights for practitioners designing privacy-aware dense retrieval systems. We also propose a simple embedding transformation fix that guarantees equal ranking effectiveness while mitigating the recoverability risk. Overall, this study reveals that Vec2Text could pose a threat to current dense retrieval systems, but there are some effective methods to patch such systems.
Authors: Ziteng Sun, Uri Mendlovic, Yaniv Leviathan, Asaf Aharoni, Ahmad Beirami, Jae Hun Ro, Ananda Theertha Suresh
Abstract: Speculative decoding is an effective method for lossless acceleration of large language models during inference. It uses a fast model to draft a block of tokens which are then verified in parallel by the target model, and provides a guarantee that the output is distributed identically to a sample from the target model. In prior works, draft verification is performed independently token-by-token. Surprisingly, we show that this approach is not optimal. We propose Block Verification, a simple draft verification algorithm that verifies the entire block jointly and provides additional wall-clock speedup. We prove that the proposed mechanism is optimal in the expected number of tokens produced each iteration and specifically is never worse than the standard token-level verification. Empirically, block verification provides modest but consistent wall-clock speedups over the standard token verification algorithm of 5%-8% in a range of tasks and datasets. Given that block verification does not increase code complexity, maintains the strong lossless guarantee of the standard speculative decoding verification algorithm, cannot deteriorate performance, and, in fact, consistently improves it, it can be used as a good default in speculative decoding implementations.
Authors: Akshat Gupta, Dev Sajnani, Gopala Anumanchipalli
Abstract: ROME and MEMIT are largely believed to be two different model editing algorithms, with the major difference between them being the ability to perform batched edits. In this paper, we unify these two algorithms under a single conceptual umbrella, optimizing for the same goal, which we call the preservation-memorization objective. ROME uses an equality constraint to optimize this objective to perform one edit at a time, whereas MEMIT employs a more flexible least-square constraint that allows for batched edits. We generalize ROME and enable batched editing with equality constraint in the form of EMMET - an Equality-constrained Mass Model Editing algorithm for Transformers, a new batched memory-editing algorithm. EMMET can perform batched-edits up to a batch-size of 10,000, with very similar performance to MEMIT across multiple dimensions. With the introduction of EMMET, we truly unify ROME and MEMIT and show that both algorithms are equivalent in terms of their optimization objective, their abilities (singular and batched editing), their model editing performance and their limitations.
Authors: Michael Hassid, Tal Remez, Jonas Gehring, Roy Schwartz, Yossi Adi
Abstract: It is a common belief that large language models (LLMs) are better than smaller-sized ones. However, larger models also require significantly more time and compute during inference. This begs the question: what happens when both models operate under the same budget? (e.g., compute, run-time). To address this question, we analyze code generation LLMs of various sizes and make comparisons such as running a 70B model once vs. generating five outputs from a 13B model. We consider a standard unit-test setup, which can be used to select the correct output from the smaller model. Our findings reveal that the repeated use of smaller models can yield consistent improvements, with gains of up to 15% across five tasks. On the other hand, in scenarios where unit-tests are unavailable, a ranking-based selection of candidates from the smaller model falls short of the performance of a single output from larger ones. Our results highlight the potential of using smaller models instead of larger ones, and the importance of studying approaches for ranking LLM outputs.
Authors: Nicholas S. Kersting, Mohammad Rahman, Suchismitha Vedala, Yang Wang
Abstract: We introduce an intuitive method to test the robustness (stability and explainability) of any black-box LLM in real-time via its local deviation from harmoniticity, denoted as $\gamma$. To the best of our knowledge this is the first completely model-agnostic and unsupervised method of measuring the robustness of any given response from an LLM, based upon the model itself conforming to a purely mathematical standard. To show general application and immediacy of results, we measure $\gamma$ in 10 popular LLMs (ChatGPT, Claude-2.1, Claude3.0, GPT-4, GPT-4o, Smaug-72B, Mixtral-8x7B, Llama2-7B, Mistral-7B and MPT-7B) across thousands of queries in three objective domains: WebQA, ProgrammingQA, and TruthfulQA. Across all models and domains tested, human annotation confirms that $\gamma \to 0$ indicates trustworthiness, and conversely searching higher values of $\gamma$ easily exposes examples of hallucination, a fact that enables efficient adversarial prompt generation through stochastic gradient ascent in $\gamma$. The low-$\gamma$ leaders among the models in the respective domains are GPT-4o, GPT-4, and Smaug-72B, providing evidence that mid-size open-source models can win out against large commercial models.
Authors: Haoran You, Yipin Guo, Yichao Fu, Wei Zhou, Huihong Shi, Xiaofan Zhang, Souvik Kundu, Amir Yazdanbakhsh, Yingyan Celine Lin
Abstract: Large language models (LLMs) have shown impressive performance on language tasks but face challenges when deployed on resource-constrained devices due to their extensive parameters and reliance on dense multiplications, resulting in high memory demands and latency bottlenecks. Shift-and-add reparameterization offers a promising solution by replacing costly multiplications with hardware-friendly primitives in both the attention and multi-layer perceptron (MLP) layers of an LLM. However, current reparameterization techniques require training from scratch or full parameter fine-tuning to restore accuracy, which is resource-intensive for LLMs. To address this, we propose accelerating pretrained LLMs through post-training shift-and-add reparameterization, creating efficient multiplication-free models, dubbed ShiftAddLLM. Specifically, we quantize each weight matrix into binary matrices paired with group-wise scaling factors. The associated multiplications are reparameterized into (1) shifts between activations and scaling factors and (2) queries and adds according to the binary matrices. To reduce accuracy loss, we present a multi-objective optimization method to minimize both weight and output activation reparameterization errors. Additionally, based on varying sensitivity across layers to reparameterization, we develop an automated bit allocation strategy to further reduce memory usage and latency. Experiments on five LLM families and eight tasks consistently validate the effectiveness of ShiftAddLLM, achieving average perplexity improvements of 5.6 and 22.7 points at comparable or lower latency compared to the most competitive quantized LLMs at 3 and 2 bits, respectively, and more than 80% memory and energy reductions over the original LLMs. Codes and models are available at https://github.com/GATECH-EIC/ShiftAddLLM.
Authors: Tomer Porian, Mitchell Wortsman, Jenia Jitsev, Ludwig Schmidt, Yair Carmon
Abstract: Kaplan et al. and Hoffmann et al. developed influential scaling laws for the optimal model size as a function of the compute budget, but these laws yield substantially different predictions. We explain the discrepancy by reproducing the Kaplan scaling law on two datasets (OpenWebText2 and RefinedWeb) and identifying three factors causing the difference: last layer computational cost, warmup duration, and scale-dependent optimizer tuning. With these factors corrected, we obtain excellent agreement with the Hoffmann et al. (i.e., "Chinchilla") scaling law. Counter to a hypothesis of Hoffmann et al., we find that careful learning rate decay is not essential for the validity of their scaling law. As a secondary result, we derive scaling laws for the optimal learning rate and batch size, finding that tuning the AdamW $\beta_2$ parameter is essential at lower batch sizes.