论文检索

Deep Learning · Large Language Models

Chengshu Li, Jacky Liang, Andy Zeng, Xinyun Chen, Karol Hausman, Dorsa Sadigh, Sergey Levine, Li Fei-Fei, Fei Xia, brian ichter

Code provides a general syntactic structure to build complex programs and perform precise computations when paired with a code interpreter – we hypothesize that language models (LMs) can leverage code-writing to improve Chain of Thought reasoning not only for logic and arithmetic tasks, but also for semantic ones (and in particular, those that are a mix of both). For example, consider prompting an LM to write code that counts the number of times it detects sarcasm in an essay: the LM may struggle to write an implementation for "detect_sarcasm(string)" that can be executed by the interpreter (handling the edge cases would be insurmountable). However, LMs may still produce a valid solution if they not only write code, but also selectively "emulate" the interpreter by generating the expected output of "detect_sarcasm(string)". In this work, we propose Chain of Code (CoC), a simple yet surprisingly effective extension that improves LM code-driven reasoning. The key idea is to encourage LMs to format semantic sub-tasks in a program as flexible pseudocode that the interpreter can explicitly catch undefined behaviors and hand off to simulate with an LM (as an "LMulator"). Experiments demonstrate that Chain of Code outperforms Chain of Thought and other baselines across a variety of benchmarks; on BIG-Bench Hard, Chain of Code achieves 84%, a gain of 12% over Chain of Thought. In a nutshell, CoC broadens the scope of reasoning questions that LMs can answer by "thinking in code".

Junyi Zou, Matthew Levine, Dessi Zaharieva, Ramesh Johari, Emily Fox

Hybrid models composing mechanistic ODE-based dynamics with flexible and expressive neural network components have grown rapidly in popularity, especially in scientific domains where such ODE-based modeling offers important interpretability and validated causal grounding (e.g., for counterfactual reasoning). The incorporation of mechanistic models also provides inductive bias in standard blackbox modeling approaches, critical when learning from small datasets or partially observed, complex systems. Unfortunately, as the hybrid models become more flexible, the causal grounding provided by the mechanistic model can quickly be lost. We address this problem by leveraging another common source of domain knowledge: *ranking* of treatment effects for a set of interventions, even if the precise treatment effect is unknown. We encode this information in a *causal loss* that we combine with the standard predictive loss to arrive at a *hybrid loss* that biases our learning towards causally valid hybrid models. We demonstrate our ability to achieve a win-win, state-of-the-art predictive performance *and* causal validity, in the challenging task of modeling glucose dynamics post-exercise in individuals with type 1 diabetes.

Miscellaneous Aspects of Machine Learning · Representation Learning

Zhengqi Pei, Anran Zhang, Shuhui Wang, Xiangyang Ji, Qingming Huang

Neural models are equivalent to dynamic systems from a physics-inspired view, implying that computation on neural networks can be interpreted as the dynamical interactions between neurons. However, existing work models neuronal interaction as a weight-based linear transformation, and the nonlinearity comes from the nonlinear activation functions, which leads to limited nonlinearity and data-fitting ability of the whole neural model. Inspired by Riemannian geometry, we interpret neural structures by projecting neurons onto the Riemannian neuronal state space and model neuronal interaction with Riemannian metric (${\it RieM}$), which provides a more efficient neural representation with higher parameter efficiency. With ${\it RieM}$, we further design a novel data-free neural compression mechanism that does not require additional fine-tuning with real data. Using backbones like ResNet and Vision Transformer, we conduct extensive experiments on datasets such as MNIST, CIFAR-100, ImageNet-1k, and COCO object detection. Empirical results show that, under equal compression rates and computational complexity, models compressed with ${\it RieM}$ achieve superior inference accuracy compared to existing data-free compression methods.

Applications · Neuroscience, Cognitive Science

Noga Mudrik, Yenho Chen, Eva Yezerets, Christopher Rozell, Adam Charles

Learning interpretable representations of neural dynamics at a population level is a crucial first step to understanding how observed neural activity relates to perception and behavior. Models of neural dynamics often focus on either low-dimensional projections of neural activity or on learning dynamical systems that explicitly relate to the neural state over time. We discuss how these two approaches are interrelated by considering dynamical systems as representative of flows on a low-dimensional manifold. Building on this concept, we propose a new decomposed dynamical system model that represents complex non-stationary and nonlinear dynamics of time series data as a sparse combination of simpler, more interpretable components. Our model is trained through a dictionary learning procedure, where we leverage recent results in tracking sparse vectors over time. The decomposed nature of the dynamics is more expressive than previous switched approaches for a given number of parameters and enables modeling of overlapping and non-stationary dynamics. In both continuous-time and discrete-time instructional examples, we demonstrate that our model effectively approximates the original system, learns efficient representations, and captures smooth transitions between dynamical modes. Furthermore, we highlight our model’s ability to efficiently capture and demix population dynamics generated from multiple independent subnetworks, a task that is computationally impractical for switched models. Finally, we apply our model to neural ``full brain'' recordings of C. elegans data, illustrating a diversity of dynamics that is obscured when classified into discrete states.

Social Aspects · Accountability, Transparency and Interpretability

Adam Stein, Aaditya Naik, Yinjun Wu, Mayur Naik, Eric Wong

Concept-based interpretability methods offer a lens into the internals of foundation models by decomposing their embeddings into high-level concepts. These concept representations are most useful when they are *compositional*, meaning that the individual concepts compose to explain the full sample. We show that existing unsupervised concept extraction methods find concepts which are not compositional. To automatically discover compositional concept representations, we identify two salient properties of such representations, and propose Compositional Concept Extraction (CCE) for finding concepts which obey these properties. We evaluate CCE on five different datasets over image and text data. Our evaluation shows that CCE finds more compositional concept representations than baselines and yields better accuracy on four downstream classification tasks.

Deep Learning · Large Language Models

Chengshu Li, Jacky Liang, Andy Zeng, Xinyun Chen, Karol Hausman, Dorsa Sadigh, Sergey Levine, Li Fei-Fei, Fei Xia, brian ichter

Code provides a general syntactic structure to build complex programs and perform precise computations when paired with a code interpreter – we hypothesize that language models (LMs) can leverage code-writing to improve Chain of Thought reasoning not only for logic and arithmetic tasks, but also for semantic ones (and in particular, those that are a mix of both). For example, consider prompting an LM to write code that counts the number of times it detects sarcasm in an essay: the LM may struggle to write an implementation for "detect_sarcasm(string)" that can be executed by the interpreter (handling the edge cases would be insurmountable). However, LMs may still produce a valid solution if they not only write code, but also selectively "emulate" the interpreter by generating the expected output of "detect_sarcasm(string)". In this work, we propose Chain of Code (CoC), a simple yet surprisingly effective extension that improves LM code-driven reasoning. The key idea is to encourage LMs to format semantic sub-tasks in a program as flexible pseudocode that the interpreter can explicitly catch undefined behaviors and hand off to simulate with an LM (as an "LMulator"). Experiments demonstrate that Chain of Code outperforms Chain of Thought and other baselines across a variety of benchmarks; on BIG-Bench Hard, Chain of Code achieves 84%, a gain of 12% over Chain of Thought. In a nutshell, CoC broadens the scope of reasoning questions that LMs can answer by "thinking in code".

Deep Learning · Generative Models and Autoencoders

Kartik Sharma, Srijan Kumar, Rakshit Trivedi

Diffusion models lend transformative capabilities to the graph generation task, yet controlling the properties of the generated graphs remains challenging. Recent approaches augment support for controlling soft, differentiable properties but they fail to handle user-specified hard constraints that are non-differentiable. This often results in vague control, unsuitable for applications like drug discovery that demand satisfaction of precise constraints, e.g., the maximum number of bonds. To address this, we formalize the problem of controlled graph generation and introduce PRODIGY (PROjected DIffusion for controlled Graph Generation), an innovative plug-and-play approach enabling the generation of graphs with precise control, from any pre-trained diffusion model. PRODIGY employs a novel operator to project the samples at each diffusion step onto the specified constrained space. For a large class of practical constraints and a variety of graphs, our extensive experiments demonstrate that PRODIGY empowers state-of-the-art continuous and discrete diffusion models to produce graphs meeting specific, hard constraints. Our approach achieves up to 100% constraint satisfaction for non-attributed and molecular graphs, under a variety of constraints, marking a significant step forward in precise, interpretable graph generation. Code is provided on the project webpage: [https://prodigy-diffusion.github.io/](https://prodigy-diffusion.github.io/).

Theory · Learning Theory

Samuel Deng, Daniel Hsu

The multi-group learning model formalizes the learning scenario in which a single predictor must generalize well on multiple, possibly overlapping subgroups of interest. We extend the study of multi-group learning to the natural case where the groups are hierarchically structured. We design an algorithm for this setting that outputs an interpretable and deterministic decision tree predictor with near-optimal sample complexity. We then conduct an empirical evaluation of our algorithm and find that it achieves attractive generalization properties on real datasets with hierarchical group structure.

Deep Learning · Generative Models and Autoencoders

Randall Balestriero, Yann LeCun

Input space reconstruction is an attractive representation learning paradigm. Despite interpretability benefit of reconstruction and generation, we identify a misalignment between learning to reconstruct, and learning for perception. We show that the former allocates a model's capacity towards a subspace of the data explaining the observed variance--a subspace with uninformative features for the latter. For example, the supervised TinyImagenet task with images projected onto the top subspace explaining 90% of the pixel variance can be solved with 45% test accuracy. Using the bottom subspace instead, accounting for only 20% of the pixel variance, reaches 55% test accuracy. Learning by reconstruction is also wasteful as the features for perception are learned last, pushing the need for long training schedules. We finally prove that learning by denoising can alleviate that misalignment for some noise strategies, e.g., masking. While tuning the noise strategy without knowledge of the perception task seems challenging, we provide a solution to detect if a noise strategy is never beneficial regardless of the perception task, e.g., additive Gaussian noise.

Miscellaneous Aspects of Machine Learning · Causality

Junyi Zou, Matthew Levine, Dessi Zaharieva, Ramesh Johari, Emily Fox

Hybrid models composing mechanistic ODE-based dynamics with flexible and expressive neural network components have grown rapidly in popularity, especially in scientific domains where such ODE-based modeling offers important interpretability and validated causal grounding (e.g., for counterfactual reasoning). The incorporation of mechanistic models also provides inductive bias in standard blackbox modeling approaches, critical when learning from small datasets or partially observed, complex systems. Unfortunately, as the hybrid models become more flexible, the causal grounding provided by the mechanistic model can quickly be lost. We address this problem by leveraging another common source of domain knowledge: *ranking* of treatment effects for a set of interventions, even if the precise treatment effect is unknown. We encode this information in a *causal loss* that we combine with the standard predictive loss to arrive at a *hybrid loss* that biases our learning towards causally valid hybrid models. We demonstrate our ability to achieve a win-win, state-of-the-art predictive performance *and* causal validity, in the challenging task of modeling glucose dynamics post-exercise in individuals with type 1 diabetes.

Probabilistic Methods

Michael Li, Emily Fox, Noah Goodman

Statistical model discovery is a challenging search over a vast space of models subject to domain-specific constraints. Efficiently searching over this space requires expertise in modeling and the problem domain. Motivated by the domain knowledge and programming capabilities of large language models (LMs), we introduce a method for language model driven automated statistical model discovery. We cast our automated procedure within the principled framework of Box’s Loop: the LM iterates between proposing statistical models represented as probabilistic programs, acting as a modeler, and critiquing those models, acting as a domain expert. By leveraging LMs, we do not have to define a domain-specific language of models or design a handcrafted search procedure, which are key restrictions of previous systems. We evaluate our method in three settings in probabilistic modeling: searching within a restricted space of models, searching over an open-ended space, and improving expert models under natural language constraints (e.g., this model should be interpretable to an ecologist). Our method identifies models on par with human expert designed models and extends classic models in interpretable ways. Our results highlight the promise of LM-driven model discovery.

Miscellaneous Aspects of Machine Learning · Online Learning, Active Learning and Bandits

Michał Zając, Tinne Tuytelaars, Gido M van de Ven

Class-incremental learning (CIL) is a particularly challenging variant of continual learning, where the goal is to learn to discriminate between all classes presented in an incremental fashion. Existing approaches often suffer from excessive forgetting and imbalance of the scores assigned to classes that have not been seen together during training. In this study, we introduce a novel approach, Prediction Error-based Classification (PEC), which differs from traditional discriminative and generative classification paradigms. PEC computes a class score by measuring the prediction error of a model trained to replicate the outputs of a frozen random neural network on data from that class. The method can be interpreted as approximating a classification rule based on Gaussian Process posterior variance. PEC offers several practical advantages, including sample efficiency, ease of tuning, and effectiveness even when data are presented one class at a time. Our empirical results show that PEC performs strongly in single-pass-through-data CIL, outperforming other rehearsal-free baselines in all cases and rehearsal-based methods with moderate replay buffer size in most cases across multiple benchmarks.

Deep Learning · Graph Neural Networks

Kelly Maggs, Celia Hacker, Bastian Rieck

Geometric deep learning extends deep learning to incorporate information about the geometry and topology data, especially in complex domains like graphs. Despite the popularity of message passing in this field, it has limitations such as the need for graph rewiring, ambiguity in interpreting data, and over-smoothing. In this paper, we take a different approach, focusing on leveraging geometric information from simplicial complexes embedded in $\mathbb{R}^n$ using node coordinates. We use differential $k$-forms in $\mathbb{R}^n$ to create representations of simplices, offering interpretability and geometric consistency without message passing. This approach also enables us to apply differential geometry tools and achieve universal approximation. Our method is efficient, versatile, and applicable to various input complexes, including graphs, simplicial complexes, and cell complexes. It outperforms existing message passing neural networks in harnessing information from geometrical graphs with node features serving as coordinates.

Applications · Robotics

Jiayuan Gu, Sean Kirmani, Paul Wohlhart, Yao Lu, Montserrat Gonzalez Arenas, Kanishka Rao, Wenhao Yu, Chuyuan Fu, Keerthana Gopalakrishnan, Zhuo Xu 等

Generalization remains one of the most important desiderata for robust robot learning systems. While recently proposed approaches show promise in generalization to novel objects, semantic concepts, or visual distribution shifts, generalization to new tasks remains challenging. For example, a language-conditioned policy trained on pick-and-place tasks will not be able to generalize to a folding task, even if the arm trajectory of folding is similar to pick-and-place. Our key insight is that this kind of generalization becomes feasible if we represent the task through rough trajectory sketches. We propose a policy conditioning method using such rough trajectory sketches, which we call RT-Trajectory, that is practical, easy to specify, and allows the policy to effectively perform new tasks that would otherwise be challenging to perform. We find that trajectory sketches strike a balance between being detailed enough to express low-level motion-centric guidance while being coarse enough to allow the learned policy to interpret the trajectory sketch in the context of situational visual observations. In addition, we show how trajectory sketches can provide a useful interface to communicate with robotic policies -- they can be specified through simple human inputs like drawings or videos, or through automated methods such as modern image-generating or waypoint-generating methods. We evaluate RT-Trajectory at scale on a variety of real-world robotic tasks, and find that RT-Trajectory is able to perform a wider range of tasks compared to language-conditioned and goal-conditioned policies, when provided the same training data.

Deep Learning · Large Language Models

Chujie Zheng, Hao Zhou, Fandong Meng, Jie Zhou, Minlie Huang

Multiple choice questions (MCQs) serve as a common yet important task format in the evaluation of large language models (LLMs). This work shows that modern LLMs are vulnerable to option position changes in MCQs due to their inherent “selection bias”, namely, they prefer to select specific option IDs as answers (like “Option A”). Through extensive empirical analyses with 20 LLMs on three benchmarks, we pinpoint that this behavioral bias primarily stems from LLMs’ token bias, where the model a priori assigns more probabilistic mass to specific option ID tokens (e.g., A/B/C/D) when predicting answers from the option IDs. To mitigate selection bias, we propose a label-free, inference-time debiasing method, called PriDe, which separates the model’s prior bias for option IDs from the overall prediction distribution. PriDe first estimates the prior by permutating option contents on a small number of test samples, and then applies the estimated prior to debias the remaining samples. We demonstrate that it achieves interpretable and transferable debiasing with high computational efficiency. We hope this work can draw broader research attention to the bias and robustness of modern LLMs.

Deep Learning · Other Representation Learning

Haozhe Chen, Junfeng Yang, Carl Vondrick, Chengzhi Mao

Large-scale pre-trained vision foundation models, such as CLIP, have become de facto backbones for various vision tasks. However, due to their black-box nature, understanding the underlying rules behind these models’ predictions and controlling model behaviors have remained open challenges. We present INViTE: a framework for INterpreting Vision Transformer’s latent tokens with Text Explanations. Given a latent token, INViTE retains its semantic information to the final layer using transformer’s local operations and retrieves the closest text for explanation. INViTE enables understanding of model visual reasoning procedure without needing additional model training or data collection. Based on the obtained interpretations, INViTE allows for model editing that controls model reasoning behaviors and improves model robustness against biases and spurious correlations. Our code is available at https://github.com/tonychenxyz/vit-interpret.

Linlu Qiu, Liwei Jiang, Ximing Lu, Melanie Sclar, Valentina Pyatkin, Chandra Bhagavatula, Bailin Wang, Yoon Kim, Yejin Choi, Nouha Dziri 等

The ability to derive underlying principles from a handful of observations and then generalize to novel situations---known as inductive reasoning---is central to human intelligence. Prior work suggests that language models (LMs) often fall short on inductive reasoning, despite achieving impressive success on research benchmarks. In this work, we conduct a systematic study of the inductive reasoning capabilities of LMs through $\textit{iterative hypothesis refinement}$, a technique that more closely mirrors the human inductive process than standard input-output prompting. Iterative hypothesis refinement employs a three-step process: proposing, selecting, and refining hypotheses in the form of textual rules. By examining the intermediate rules, we observe that LMs are phenomenal $\textit{hypothesis proposers}$ (i.e., generating candidate rules), and when coupled with a (task-specific) symbolic interpreter that is able to systematically filter the proposed set of rules, this hybrid approach achieves strong results across inductive reasoning benchmarks that require inducing causal relations, language-like instructions, and symbolic concepts. However, they also behave as puzzling $\textit{inductive reasoners}$, showing notable performance gaps between rule induction (i.e., identifying plausible rules) and rule application (i.e., applying proposed rules to instances), suggesting that LMs are proposing hypotheses without being able to actually apply the rules. Through empirical and human analyses, we further reveal several discrepancies between the inductive reasoning processes of LMs and humans, shedding light on both the potentials and limitations of using LMs in inductive reasoning tasks.

Deep Learning · Large Language Models

Linlu Qiu, Liwei Jiang, Ximing Lu, Melanie Sclar, Valentina Pyatkin, Chandra Bhagavatula, Bailin Wang, Yoon Kim, Yejin Choi, Nouha Dziri 等

The ability to derive underlying principles from a handful of observations and then generalize to novel situations---known as inductive reasoning---is central to human intelligence. Prior work suggests that language models (LMs) often fall short on inductive reasoning, despite achieving impressive success on research benchmarks. In this work, we conduct a systematic study of the inductive reasoning capabilities of LMs through $\textit{iterative hypothesis refinement}$, a technique that more closely mirrors the human inductive process than standard input-output prompting. Iterative hypothesis refinement employs a three-step process: proposing, selecting, and refining hypotheses in the form of textual rules. By examining the intermediate rules, we observe that LMs are phenomenal $\textit{hypothesis proposers}$ (i.e., generating candidate rules), and when coupled with a (task-specific) symbolic interpreter that is able to systematically filter the proposed set of rules, this hybrid approach achieves strong results across inductive reasoning benchmarks that require inducing causal relations, language-like instructions, and symbolic concepts. However, they also behave as puzzling $\textit{inductive reasoners}$, showing notable performance gaps between rule induction (i.e., identifying plausible rules) and rule application (i.e., applying proposed rules to instances), suggesting that LMs are proposing hypotheses without being able to actually apply the rules. Through empirical and human analyses, we further reveal several discrepancies between the inductive reasoning processes of LMs and humans, shedding light on both the potentials and limitations of using LMs in inductive reasoning tasks.

Deep Learning · Large Language Models

Bill Yuchen Lin, Abhilasha Ravichander, Ximing Lu, Nouha Dziri, Melanie Sclar, Khyathi Chandu, Chandra Bhagavatula, Yejin Choi

Alignment tuning has become the de facto standard practice for enabling base large language models (LLMs) to serve as open-domain AI assistants. The alignment tuning process typically involves instruction learning through supervised fine-tuning (SFT) and preference tuning via reinforcement learning from human feedback (RLHF). A recent study, LIMA (Zhou et al., 2023), shows that using merely 1K examples for SFT can achieve significant alignment performance as well, suggesting that the effect of alignment tuning might be "superficial." This raises questions about how exactly the alignment tuning transforms a base LLM. We analyze the effect of alignment tuning by examining the token distribution shift between base LLMs and their aligned counterparts (e.g., Llama-2 and Llama-2-chat). Our findings reveal that base LLMs and their alignment-tuned versions perform nearly identically in decoding on the majority of token positions (i.e., they share the top-ranked tokens). Most distribution shifts occur with stylistic tokens (e.g., discourse markers, safety disclaimers). This direct evidence strongly supports the hypothesis that alignment tuning primarily learns to adopt the language style of AI assistants, and that the knowledge required for answering user queries predominantly comes from the base LLMs themselves. Based on these findings, we rethink the alignment of LLMs by posing the research question: how effectively can we align base LLMs without SFT or RLHF? To address this, we introduce a simple, tuning-free alignment method, URIAL (Untuned LLMs with Restyled In-context Alignment). URIAL achieves effective alignment purely through in-context learning (ICL) with base LLMs, requiring as few as three constant stylistic examples and a system prompt. We conduct a fine-grained and interpretable evaluation on a diverse set of examples, named just-eval-instruct. Results demonstrate that base LLMs with URIAL can match or even surpass the performance of LLMs aligned with SFT (Mistral-7b-Instruct) or SFT+RLHF (Llama-2-70b-chat). We show that the gap between tuning-free and tuning-based alignment methods can be significantly reduced through strategic prompting and ICL. Our findings on the superficial nature of alignment tuning and results with URIAL suggest that deeper analysis and theoretical understanding of alignment is crucial to future LLM research.

Deep Learning · Generative Models and Autoencoders

Peter West, Ximing Lu, Nouha Dziri, Faeze Brahman, Linjie Li, Jena Hwang, Liwei Jiang, Jillian Fisher, Abhilasha Ravichander, Khyathi Chandu 等

The recent wave of generative AI has sparked unprecedented global attention, with both excitement and concern over potentially superhuman levels of artificial intelligence: models now take only seconds to produce outputs that would challenge or exceed the capabilities even of expert humans. At the same time, models still show basic errors in understanding that would not be expected even in non-expert humans. This presents us with an apparent paradox: how do we reconcile seemingly superhuman capabilities with the persistence of errors that few humans would make? In this work, we posit that this tension reflects a divergence in the configuration of intelligence in today's generative models relative to intelligence in humans. Specifically, we propose and test the **Generative AI Paradox** hypothesis: generative models, having been trained directly to reproduce expert-like outputs, acquire generative capabilities that are not contingent upon---and can therefore exceed---their ability to understand those same types of outputs. This contrasts with humans, for whom basic understanding almost always precedes the ability togenerate expert-level outputs. We test this hypothesis through controlled experiments analyzing generation vs.~understanding in generative models, across both language and image modalities. Our results show that although models can outperform humans in generation, they consistently fall short of human capabilities in measures of understanding, as well as weaker correlation between generation and understanding performance, and more brittleness to adversarial inputs. Our findings support the hypothesis that models' generative capability may not be contingent upon understanding capability, and call for caution in interpreting artificial intelligence by analogy to human intelligence.