MMCOMPOSITION: Revisiting the Compositionality of Pre-trained Vision-Language Models

The advent of large Vision-Language Models (VLMs) has significantly advanced multimodal understanding, enabling more sophisticated and accurate integration of visual and textual information across various tasks, including image and video captioning, visual question answering, and cross-modal retrieval. Despite VLMs' superior capabilities, researchers lack a comprehensive understanding of their compositionality -- the ability to understand and produce novel combinations of known visual and textual components. Prior benchmarks provide only a relatively rough compositionality evaluation from the perspectives of objects, relations, and attributes while neglecting deeper reasoning about object interactions, counting, and complex compositions. However, compositionality is a critical ability that facilitates coherent reasoning and understanding across modalities for VLMs. To address this limitation, we propose MMCOMPOSITION, a novel human-annotated benchmark for comprehensively and accurately evaluating VLMs' compositionality. Our proposed benchmark serves as a complement to these earlier works. With MMCOMPOSITION, we can quantify and explore the compositionality of the mainstream VLMs. Surprisingly, we find GPT-4o's compositionality inferior to the best open-source model, and we analyze the underlying reasons. Our experimental analysis reveals the limitations of VLMs in fine-grained compositional perception and reasoning, and points to areas for improvement in VLM design and training. Resources available at: https://hanghuacs.github.io/MMComposition/

A Law of Next-Token Prediction in Large Language Models

Large language models (LLMs) have been widely employed across various application domains, yet their black-box nature poses significant challenges to understanding how these models process input data internally to make predictions. In this paper, we introduce a precise and quantitative law that governs the learning of contextualized token embeddings through intermediate layers in pre-trained LLMs for next-token prediction. Our findings reveal that each layer contributes equally to enhancing prediction accuracy, from the lowest to the highest layer -- a universal phenomenon observed across a diverse array of open-source LLMs, built on architectures such as Transformer, RWKV, and Mamba. We demonstrate that this law offers new perspectives and insights to inform and guide practices in LLM development and applications, including model scaling, pre-training tasks, and information flow. Overall, our law enables more fine-grained approaches to the design, training, and interpretation of LLMs through scrutinizing their internal data processing mechanisms.

An Empirical Analysis on Large Language Models in Debate Evaluation

In this study, we investigate the capabilities and inherent biases of advanced large language models (LLMs) such as GPT-3.5 and GPT-4 in the context of debate evaluation. We discover that LLM's performance exceeds humans and surpasses the performance of state-of-the-art methods fine-tuned on extensive datasets in debate evaluation. We additionally explore and analyze biases present in LLMs, including positional bias, lexical bias, order bias, which may affect their evaluative judgments. Our findings reveal a consistent bias in both GPT-3.5 and GPT-4 towards the second candidate response presented, attributed to prompt design. We also uncover lexical biases in both GPT-3.5 and GPT-4, especially when label sets carry connotations such as numerical or sequential, highlighting the critical need for careful label verbalizer selection in prompt design. Additionally, our analysis indicates a tendency of both models to favor the debate's concluding side as the winner, suggesting an end-of-discussion bias.

Unveiling Divergent Inductive Biases of LLMs on Temporal Data

Unraveling the intricate details of events in natural language necessitates a subtle understanding of temporal dynamics. Despite the adeptness of Large Language Models (LLMs) in discerning patterns and relationships from data, their inherent comprehension of temporal dynamics remains a formidable challenge. This research meticulously explores these intrinsic challenges within LLMs, with a specific emphasis on evaluating the performance of GPT-3.5 and GPT-4 models in the analysis of temporal data. Employing two distinct prompt types, namely Question Answering (QA) format and Textual Entailment (TE) format, our analysis probes into both implicit and explicit events. The findings underscore noteworthy trends, revealing disparities in the performance of GPT-3.5 and GPT-4. Notably, biases toward specific temporal relationships come to light, with GPT-3.5 demonstrating a preference for "AFTER'' in the QA format for both implicit and explicit events, while GPT-4 leans towards "BEFORE''. Furthermore, a consistent pattern surfaces wherein GPT-3.5 tends towards "TRUE'', and GPT-4 exhibits a preference for "FALSE'' in the TE format for both implicit and explicit events. This persistent discrepancy between GPT-3.5 and GPT-4 in handling temporal data highlights the intricate nature of inductive bias in LLMs, suggesting that the evolution of these models may not merely mitigate bias but may introduce new layers of complexity.

SocREval: Large Language Models with the Socratic Method for Reference-Free Reasoning Evaluation

To comprehensively assess the capacity of current models for complex reasoning, it is crucial to assess their step-by-step reasoning in a scalable manner. Established reference-based evaluation metrics rely on human-annotated reasoning chains to assess the model-derived chains. However, such ``gold-standard'' human-written reasoning chains may not be unique and their acquisition is often labor-intensive. Existing reference-free reasoning metrics eliminate the need for human-crafted reasoning chains as references, but they typically require fine-tuning on datasets with human-derived reasoning chains, which complicates the process and raises concerns regarding generalizability across diverse datasets. To address these challenges, we harness GPT-4 to automatically evaluate reasoning chain quality, obviating the need for human-crafted references. Leveraging the Socratic method, we devise tailored prompts to enhance reference-free reasoning evaluation, which we term SocREval (Socratic method for Reasoning Evaluation). Empirical results from four human annotated datasets reveal that SocREval significantly improves GPT-4's performance, surpassing existing reference-free and reference-based reasoning evaluation metrics. Beyond its demonstrated efficacy, our proposed framework, large language models (LLMs) with the Socratic method, proves to be both cost-efficient and robust to prompt writing and example selection, as substantiated by our in-depth analysis.

On Regularization and Inference with Label Constraints

Prior knowledge and symbolic rules in machine learning are often expressed in the form of label constraints, especially in structured prediction problems. In this work, we compare two common strategies for encoding label constraints in a machine learning pipeline, regularization with constraints and constrained inference, by quantifying their impact on model performance. For regularization, we show that it narrows the generalization gap by precluding models that are inconsistent with the constraints. However, its preference for small violations introduces a bias toward a suboptimal model. For constrained inference, we show that it reduces the population risk by correcting a model's violation, and hence turns the violation into an advantage. Given these differences, we further explore the use of two approaches together and propose conditions for constrained inference to compensate for the bias introduced by regularization, aiming to improve both the model complexity and optimal risk.

Rethinking with Retrieval: Faithful Large Language Model Inference

Despite the success of large language models (LLMs) in various natural language processing (NLP) tasks, the stored knowledge in these models may inevitably be incomplete, out-of-date, or incorrect. This motivates the need to utilize external knowledge to assist LLMs. Unfortunately, current methods for incorporating external knowledge often require additional training or fine-tuning, which can be costly and may not be feasible for LLMs. To address this issue, we propose a novel post-processing approach, rethinking with retrieval (RR), which retrieves relevant external knowledge based on the decomposed reasoning steps obtained from the chain-of-thought (CoT) prompting. This lightweight approach does not require additional training or fine-tuning and is not limited by the input length of LLMs. We evaluate the effectiveness of RR through extensive experiments with GPT-3 on three complex reasoning tasks: commonsense reasoning, temporal reasoning, and tabular reasoning. Our results show that RR can produce more faithful explanations and improve the performance of LLMs.

A Law of Data Separation in Deep Learning

Multilayer neural networks have achieved superhuman performance in many artificial intelligence applications. However, their black-box nature obscures the underlying mechanism for transforming input data into labels throughout all layers, thus hindering architecture design for new tasks and interpretation for high-stakes decision makings. We addressed this problem by introducing a precise law that governs how real-world deep neural networks separate data according to their class membership from the bottom layers to the top layers in classification problems. This law shows that each layer roughly improves a certain measure of data separation by an \textit{equal} multiplicative factor. This law manifests in modern architectures such as AlexNet, VGGNet, and ResNet in the late phase of training. This law together with the perspective of data separation offers practical guidelines for designing network architectures, improving model robustness and out-of-sample performance during training, as well as interpreting deep learning predictions.

Weighted Training for Cross-Task Learning

In this paper, we introduce Target-Aware Weighted Training (TAWT), a weighted training algorithm for cross-task learning based on minimizing a representation-based task distance between the source and target tasks. We show that TAWT is easy to implement, is computationally efficient, requires little hyperparameter tuning, and enjoys non-asymptotic learning-theoretic guarantees. The effectiveness of TAWT is corroborated through extensive experiments with BERT on four sequence tagging tasks in natural language processing (NLP), including part-of-speech (PoS) tagging, chunking, predicate detection, and named entity recognition (NER). As a byproduct, the proposed representation-based task distance allows one to reason in a theoretically principled way about several critical aspects of cross-task learning, such as the choice of the source data and the impact of fine-tuning

Layer-Peeled Model: Toward Understanding Well-Trained Deep Neural Networks

In this paper, we introduce the Layer-Peeled Model, a nonconvex yet analytically tractable optimization program, in a quest to better understand deep neural networks that are trained for a sufficiently long time. As the name suggests, this new model is derived by isolating the topmost layer from the remainder of the neural network, followed by imposing certain constraints separately on the two parts. We demonstrate that the Layer-Peeled Model, albeit simple, inherits many characteristics of well-trained neural networks, thereby offering an effective tool for explaining and predicting common empirical patterns of deep learning training. First, when working on class-balanced datasets, we prove that any solution to this model forms a simplex equiangular tight frame, which in part explains the recently discovered phenomenon of neural collapse in deep learning training [PHD20]. Moreover, when moving to the imbalanced case, our analysis of the Layer-Peeled Model reveals a hitherto unknown phenomenon that we term Minority Collapse, which fundamentally limits the performance of deep learning models on the minority classes. In addition, we use the Layer-Peeled Model to gain insights into how to mitigate Minority Collapse. Interestingly, this phenomenon is first predicted by the Layer-Peeled Model before its confirmation by our computational experiments.