Catastrophic forgetting (CF) poses a significant challenge in machine learning, where a model forgets previously learned information upon learning new tasks. Despite the advanced capabilities of Large Language Models (LLMs), they continue to face challenges with CF during continual learning. The majority of existing research focuses on analyzing forgetting patterns through a singular training sequence, thereby overlooking the intricate effects that diverse tasks have on model behavior.Our study explores CF across various settings, discovering that model forgetting is influenced by both the specific training tasks and the models themselves. To this end, we interpret forgetting by examining the function vector (FV), a compact representation of functions in LLMs, offering a model-dependent indicator for the occurrence of CF. Through theoretical and empirical analyses, we demonstrated that CF in LLMs primarily stems from biases in function activation rather than the overwriting of task processing functions.Leveraging these insights, we propose a novel function vector guided training methodology, incorporating a regularization technique to stabilize the FV and mitigate forgetting. Empirical tests on four benchmarks confirm the effectiveness of our proposed training method, substantiating our theoretical framework concerning CF and model function dynamics.
论文检索
输入标题、作者或关键词，从 7,537 篇学术成果中精准定位
Deep Learning · Large Language Models
Catastrophic forgetting (CF) poses a significant challenge in machine learning, where a model forgets previously learned information upon learning new tasks. Despite the advanced capabilities of Large Language Models (LLMs), they continue to face challenges with CF during continual learning. The majority of existing research focuses on analyzing forgetting patterns through a singular training sequence, thereby overlooking the intricate effects that diverse tasks have on model behavior.Our study explores CF across various settings, discovering that model forgetting is influenced by both the specific training tasks and the models themselves. To this end, we interpret forgetting by examining the function vector (FV), a compact representation of functions in LLMs, offering a model-dependent indicator for the occurrence of CF. Through theoretical and empirical analyses, we demonstrated that CF in LLMs primarily stems from biases in function activation rather than the overwriting of task processing functions.Leveraging these insights, we propose a novel function vector guided training methodology, incorporating a regularization technique to stabilize the FV and mitigate forgetting. Empirical tests on four benchmarks confirm the effectiveness of our proposed training method, substantiating our theoretical framework concerning CF and model function dynamics.
Deep Learning · Other Representation Learning
Counting constitutes a core skill underlying a wide range of tasks, such as formal language recognition, multi-hop reasoning and simulating algorithms. Generaliz- ing counting inductively is central to task success on out-of-distribution (OOD) instances where testing inputs are longer than those seen in training. While there is a large body of literature reporting poor length generalization in language models, few papers have tried to distill the “reasoning” failure to the simplest case of count- ing failure. We aim to provide a broader picture on whether various language model architectures can a) learn to count, and b) generalize counting inductively. This work provides extensive empirical results on architectures ranging from RNNs, Transformers, State-Space Models and RWKV. We present carefully-designed task formats, auxiliary tasks and positional embeddings to avoid limitations in general- ization with OOD-position and OOD-vocabulary. We find that while traditional RNNs trivially achieve inductive counting, Transformers have to rely on positional embeddings (PEs) to count OOD. Further analyses on interpreting the learned solution reveal that different PEs encode different inductive biases that facilitate counting in different task formats. As counting is the basis for many arguments concerning the expressivity of Transformers, our finding calls for the community to reexamine the application scope of primitive functions defined in formal charac- terizations. Finally, modern RNNs also largely underperform traditional RNNs in generalizing counting inductively, hinting at the tradeoff modern RNNs struggle to balance between parallelized training and maintaining their recurrent nature.
Applications · Computer Vision
In the realm of image generation, creating customized images from visual prompt with additional textual instruction emerges as a promising endeavor. However, existing methods, both tuning-based and tuning-free, struggle with interpreting the subject-essential attributes from the visual prompt. This leads to subject-irrelevant attributes infiltrating the generation process, ultimately compromising the personalization quality in both editability and ID preservation. In this paper, we present $\textbf{DisEnvisioner}$, a novel approach for effectively extracting and enriching the subject-essential features while filtering out -irrelevant information, enabling exceptional customization performance, in a $\textbf{tuning-free}$ manner and using only $\textbf{a single image}$. Specifically, the feature of the subject and other irrelevant components are effectively separated into distinctive visual tokens, enabling a much more accurate customization. Aiming to further improving the ID consistency, we enrich the disentangled features, sculpting them into a more granular representation. Experiments demonstrate the superiority of our approach over existing methods in instruction response (editability), ID consistency, inference speed, and the overall image quality, highlighting the effectiveness and efficiency of DisEnvisioner.
Deep Learning · Graph Neural Networks
Among the many variants of graph neural network (GNN) architectures capable of modeling data with cross-instance relations, an important subclass involves layers designed such that the forward pass iteratively reduces a graph-regularized energy function of interest. In this way, node embeddings produced at the output layer dually serve as both predictive features for solving downstream tasks (e.g., node classification) and energy function minimizers that inherit transparent, exploitable inductive biases and interpretability. However, scaling GNN architectures constructed in this way remains challenging, in part because the convergence of the forward pass may involve models with considerable depth. To tackle this limitation, we propose a sampling-based energy function and scalable GNN layers that iteratively reduce it, guided by convergence guarantees in certain settings. We also instantiate a full GNN architecture based on these designs, and the model achieves competitive accuracy and scalability when applied to the largest publicly-available node classification benchmark exceeding 1TB in size. Our source code is available at https://github.com/haitian-jiang/MuseGNN.
Social Aspects · Accountability, Transparency and Interpretability
How do multi-modal generative models describe images of recent historical events and figures, whose legacies may be nuanced, multifaceted, or contested? This task necessitates not only accurate visual recognition, but also socio-cultural knowledge and cross-modal reasoning. To address this evaluation challenge, we introduce Century -- a novel dataset of sensitive historical images. This dataset consists of 1,500 images from recent history, created through an automated method combining knowledge graphs and language models with quality and diversity criteria created from the practices of museums and digital archives. We demonstrate through automated and human evaluation that this method produces a set of images that depict events and figures that are diverse across topics and represents all regions of the world.We additionally propose an evaluation framework for evaluating the historical contextualisation capabilities along dimensions of accuracy, thoroughness, and objectivity. We demonstrate this approach by using Century to evaluate four foundation models, scoring performance using both automated and human evaluation. We find that historical contextualisation of sensitive images poses a significant challenge for modern multi-modal foundation models, and offer practical recommendations for how developers can use Century to evaluate improvements to models and applications.
Mechanistic interpretability has made great strides in identifying neural network features (e.g., directions in hidden activation space) that mediate concepts (e.g., *the birth year of a Nobel laureate*) and enable predictable manipulation. Distributed alignment search (DAS) leverages supervision from counterfactual data to learn concept features within hidden states, but DAS assumes we can afford to conduct a brute force search over potential feature locations. To address this, we present HyperDAS, a transformer-based hypernetwork architecture that (1) automatically locates the token-positions of the residual stream that a concept is realized in and (2) learns features of those residual stream vectors for the concept. In experiments with Llama3-8B, HyperDAS achieves state-of-the-art performance on the RAVEL benchmark for disentangling concepts in hidden states. In addition, we review the design decisions we made to mitigate the concern that HyperDAS (like all powerful interpretabilty methods) might inject new information into the target model rather than faithfully interpreting it.
Applications · Language, Speech and Dialog
Foundation models, such as large language models, have achieved remarkable success in natural language processing and are evolving into models capable of handling multiple modalities.Listening ability, in particular, is crucial for many applications, leading to research on building speech foundation models. However, the high computational cost of these large models presents a significant challenge for real-world applications. Although substantial efforts have been made to reduce computational costs, such as through pruning techniques, the majority of these approaches are applied primarily during the training phase for specific downstream tasks. In this study, we hypothesize that optimal pruned networks may vary based on contextual factors such as speaker characteristics, languages, and tasks. To address this, we propose a dynamic pruning technique that adapts to these contexts during inference without altering the underlying model. We demonstrated that we could successfully reduce inference time by approximately 30\% while maintaining accuracy in multilingual/multi-task scenarios. We also found that the obtained pruned structure offers meaningful interpretations based on the context, e.g., task-related information emerging as the dominant factor for efficient pruning.
Applications · Chemistry and Drug Discovery
Molecular relational learning (MRL) seeks to understand the interaction behaviors between molecules, a pivotal task in domains such as drug discovery and materials science. Recently, extracting core substructures and modeling their interactions have emerged as mainstream approaches within machine learning-assisted methods. However, these methods still exhibit some limitations, such as insufficient consideration of molecular interactions or capturing substructures that include excessive noise, which hampers precise core substructure extraction.To address these challenges, we present an integrated dynamic framework called Iterative Substructure Extraction (ISE). ISE employs the Expectation-Maximization (EM) algorithm for MRL tasks, where the core substructures of interacting molecules are treated as latent variables and model parameters, respectively. Through iterative refinement, ISE gradually narrows the interactions from the entire molecular structures to just the core substructures.Moreover, to ensure the extracted substructures are concise and compact, we propose the Interactive Graph Information Bottleneck (IGIB) theory, which focuses on capturing the most influential yet minimal interactive substructures. In summary, our approach, guided by the IGIB theory, achieves precise substructure extraction within the ISE framework and is encapsulated in the IGIB-ISE}Extensive experiments validate the superiority of our model over state-of-the-art baselines across various tasks in terms of accuracy, generalizability, and interpretability.
Social Aspects · Accountability, Transparency and Interpretability
Automated mechanistic interpretation research has attracted great interest due to its potential to scale explanations of neural network internals to large models. Existing automated circuit discovery work relies on activation patching or its approximations to identify subgraphs in models for specific tasks (circuits). They often suffer from slow runtime, approximation errors, and specific requirements of metrics, such as non-zero gradients.In this work, we introduce contextual decomposition for transformers (CD-T) to build interpretable circuits in large language models. CD-T can produce circuits at any level of abstraction and is the first to efficiently produce circuits as fine-grained as attention heads at specific sequence positions.CD-T is compatible to all transformer types, and requires no training or manually-crafted examples.CD-T consists of a set of mathematical equations to isolate contribution of model features. Through recursively computing contribution of all nodes in a computational graph of a model using CD-T followed by pruning, we are able to reduce circuit discovery runtime from hours to seconds compared to state-of-the-art baselines.On three standard circuit evaluation datasets (indirect object identification, greater-than comparisons, and docstring completion),we demonstrate that CD-T outperforms ACDC and EAP by better recovering the manual circuits with an average of 97% ROC AUC under low runtimes.In addition, we provide evidence that faithfulness of CD-T circuits is not due to random chance by showing our circuits are 80% more faithful than random circuits of up to 60% of the original model size. Finally, we show CD-T circuits are able to perfectly replicate original models' behavior(faithfulness = 1) using fewer nodes than the baselines for all tasks.Our results underscore the great promise of CD-T for efficient automated mechanistic interpretability, paving the way for new insights into the workings of large language models.
Applications · Chemistry and Drug Discovery
Generating novel crystalline materials has potential to lead to advancements in fields such as electronics, energy storage, and catalysis. The defining characteristic of crystals is their symmetry, which plays a central role in determining their physical properties. However, existing crystal generation methods either fail to generate materials that display the symmetries of real-world crystals, or simply replicate the symmetry information from examples in a database. To address this limitation, we propose SymmCD, a novel diffusion-based generative model that explicitly incorporates crystallographic symmetry into the generative process. We decompose crystals into two components and learn their joint distribution through diffusion: 1) the asymmetric unit, the smallest subset of the crystal which can generate the whole crystal through symmetry transformations, and; 2) the symmetry transformations needed to be applied to each atom in the asymmetric unit. We also use a novel and interpretable representation for these transformations, enabling generalization across different crystallographic symmetry groups. We showcase the competitive performance of SymmCD on a subset of the Materials Project, obtaining diverse and valid crystals with realistic symmetries and predicted properties.
Reinforcement Learning · Everything Else
Programmatic reinforcement learning (PRL) has been explored for representing policies through programs as a means to achieve interpretability and generalization. Despite promising outcomes, current state-of-the-art PRL methods are hindered by sample inefficiency, necessitating tens of millions of program-environment interactions. To tackle this challenge, we introduce a novel LLM-guided search framework (LLM-GS). Our key insight is to leverage the programming expertise and common sense reasoning of LLMs to enhance the efficiency of assumption-free, random-guessing search methods. We address the challenge of LLMs' inability to generate precise and grammatically correct programs in domain-specific languages (DSLs) by proposing a Pythonic-DSL strategy — an LLM is instructed to initially generate Python codes and then convert them into DSL programs. To further optimize the LLM-generated programs, we develop a search algorithm named Scheduled Hill Climbing, designed to efficiently explore the programmatic search space to improve the programs consistently. Experimental results in the Karel domain demonstrate our LLM-GS framework's superior effectiveness and efficiency. Extensive ablation studies further verify the critical role of our Pythonic-DSL strategy and Scheduled Hill Climbing algorithm. Moreover, we conduct experiments with two novel tasks, showing that LLM-GS enables users without programming skills and knowledge of the domain or DSL to describe the tasks in natural language to obtain performant programs.
Applications · Computer Vision
Recently, with the development of Neural Radiance Fields and Gaussian Splatting, 3D reconstruction techniques have achieved remarkably high fidelity. However, the latent representations learnt by these methods are highly entangled and lack interpretability. In this paper, we propose a novel part-aware compositional reconstruction method, called GaussianBlock, that enables semantically coherent and disentangled representations, allowing for precise and physical editing akin to building blocks, while simultaneously maintaining high fidelity.Our GaussianBlock introduces a hybrid representation that leverages the advantages of both primitives, known for their flexible actionability and editability, and 3D Gaussians, which excel in reconstruction quality. Specifically, we achieve semantically coherent primitives through a novel attention-guided centering loss derived from 2D semantic priors, complemented by a dynamic splitting and fusion strategy. Furthermore, we utilize 3D Gaussians that hybridize with primitives to refine structural details and enhance fidelity. Additionally, a binding inheritance strategy is employed to strengthen and maintain the connection between the two. Our reconstructed scenes are evidenced to be disentangled, compositional, and compact across diverse benchmarks, enabling seamless, direct and precise editing while maintaining high quality.
Social Aspects · Accountability, Transparency and Interpretability
In this study, we propose a novel rule-based interpretable choice model, {\bf Logic-Logit}, designed to effectively learn and explain human choices. Choice models have been widely applied across various domains—such as commercial demand forecasting, recommendation systems, and consumer behavior analysis—typically categorized as parametric, nonparametric, or deep network-based. While recent innovations have favored neural network approaches for their computational power, these flexible models often involve large parameter sets and lack interpretability, limiting their effectiveness in contexts where transparency is essential.Previous empirical evidence shows that individuals usually use {\it heuristic decision rules} to form their consideration sets, from which they then choose. These rules are often represented as {\it disjunctions of conjunctions} (i.e., OR-of-ANDs). These rules-driven, {\it consider-then-choose} decision processes enable people to quickly screen numerous alternatives while reducing cognitive and search costs. Motivated by this insight, our approach leverages logic rules to elucidate human choices, providing a fresh perspective on preference modeling. We introduce a unique combination of column generation techniques and the Frank-Wolfe algorithm to facilitate efficient rule extraction for preference modeling—a process recognized as NP-hard. Our empirical evaluation, conducted on both synthetic datasets and real-world data from commercial and healthcare domains, demonstrates that Logic-Logit significantly outperforms baseline models in terms of interpretability and accuracy.
Miscellaneous Aspects of Machine Learning · General Machine Learning Techniques
Invariant risk minimization is an important general machine learning framework that has recently been interpreted as a total variation model (IRM-TV). However, how to improve out-of-distribution (OOD) generalization in the IRM-TV setting remains unsolved. In this paper, we extend IRM-TV to a Lagrangian multiplier model named OOD-TV-IRM. We find that the autonomous TV penalty hyperparameter is exactly the Lagrangian multiplier. Thus OOD-TV-IRM is essentially a primal-dual optimization model, where the primal optimization minimizes the entire invariant risk and the dual optimization strengthens the TV penalty. The objective is to reach a semi-Nash equilibrium where the balance between the training loss and OOD generalization is maintained. We also develop a convergent primal-dual algorithm that facilitates an adversarial learning scheme. Experimental results show that OOD-TV-IRM outperforms IRM-TV in most situations.
Social Aspects · Accountability, Transparency and Interpretability
*Minimal sufficient reasons* represent a prevalent form of explanation - the smallest subset of input features which, when held constant at their corresponding values, ensure that the prediction remains unchanged. Previous *post-hoc* methods attempt to obtain such explanations but face two main limitations: (1) Obtaining these subsets poses a computational challenge, leading most scalable methods to converge towards suboptimal, less meaningful subsets; (2) These methods heavily rely on sampling out-of-distribution input assignments, potentially resulting in counterintuitive behaviors. To tackle these limitations, we propose in this work a self-supervised training approach, which we term *sufficient subset training* (SST). Using SST, we train models to generate concise sufficient reasons for their predictions as an integral part of their output. Our results indicate that our framework produces succinct and faithful subsets substantially more efficiently than competing post-hoc methods while maintaining comparable predictive performance.
Miscellaneous Aspects of Machine Learning · Representation Learning
Broadly intelligent agents should form task-specific abstractions that selectively expose the essential elements of a task, while abstracting away the complexity of the raw sensorimotor space. In this work, we present Neuro-Symbolic Predicates, a first-order abstraction language that combines the strengths of symbolic and neural knowledge representations. We outline an online algorithm for inventing such predicates and learning abstract world models. We compare our approach to hierarchical reinforcement learning, vision-language model planning, and symbolic predicate invention approaches, on both in- and out-of-distribution tasks across five simulated robotic domains. Results show that our approach offers better sample complexity, stronger out-of-distribution generalization, and improved interpretability.
Applications · Everything Else
Compositional reasoning in Vision-Language Models (VLMs) remains challenging as these models often struggle to relate objects, attributes, and spatial relationships. Recent methods aim to address these limitations by relying on the semantics of the textual description, using Large Language Models (LLMs) to break them down into subsets of questions and answers. However, these methods primarily operate on the surface level, failing to incorporate deeper lexical understanding while introducing incorrect assumptions generated by the LLM. In response to these issues, we present Caption Expansion with Contradictions and Entailments (CECE), a principled approach that leverages Natural Language Inference (NLI) to generate entailments and contradictions from a given premise. CECE produces lexically diverse sentences while maintaining their core meaning. Through extensive experiments, we show that CECE enhances interpretability and reduces overreliance on biased or superficial features. By balancing CECE along the original premise, we achieve significant improvements over previous methods without requiring additional fine-tuning, producing state-of-the-art results on benchmarks that score agreement with human judgments for image-text alignment, and achieving an increase in performance on Winoground of $+19.2\%$ (group score) and $+12.9\%$ on EqBen (group score) over the best prior work (finetuned with targeted data).
Social Aspects · Accountability, Transparency and Interpretability
Instruction-following is crucial for building AI agents with large language models (LLMs), as these models must adhere strictly to user-provided constraints and guidelines. However, LLMs often fail to follow even simple and clear instructions.To improve instruction-following behavior and prevent undesirable outputs, a deeper understanding of how LLMs' internal states relate to these outcomes is required.In this work, we investigate whether LLMs encode information in their representations that correlates with instruction-following success—a property we term ``knowing internally''.Our analysis identifies a direction in the input embedding space, termed the instruction-following dimension, that predicts whether a response will comply with a given instruction.We find that this dimension generalizes well across unseen tasks but not across unseen instruction types.We demonstrate that modifying representations along this dimension improves instruction-following success rates compared to random changes, without compromising response quality.Further investigation reveals that this dimension is more closely related to the phrasing of prompts rather than the inherent difficulty of the task or instructions. This discovery also suggests explanations for why LLMs sometimes fail to follow clear instructions and why prompt engineering is often effective, even when the content remains largely unchanged. This work provides insight into the internal workings of LLMs' instruction-following, paving the way for reliable LLM agents.
When solving challenging problems, language models (LMs) are able to identify relevant information from long and complicated contexts. To study how LMs solve retrieval tasks in diverse situations, we introduce ORION, a collection of structured retrieval tasks spanning six domains, from text understanding to coding. Each task in ORION can be represented abstractly by a request (e.g. a question) that retrieves an attribute (e.g. the character name) from a context (e.g. a story). We apply causal analysis on 18 open-source language models with sizes ranging from 125 million to 70 billion parameters. We find that LMs internally decompose retrieval tasks in a modular way: middle layers at the last token position process the request, while late layers retrieve the correct entity from the context. After causally enforcing this decomposition, models are still able to solve the original task, preserving 70% of the original correct token probability in 98 of the 106 studied model-task pairs. We connect our macroscopic decomposition with a microscopic description by performing a fine-grained case study of a question-answering task on Pythia-2.8b. Building on our high-level understanding, we demonstrate a proof of concept application for scalable internal oversight of LMs to mitigate prompt-injection while requiring human supervision on only a single input. Our solution improves accuracy drastically (from 15.5% to 97.5% on Pythia-12b). This work presents evidence of a universal emergent modular processing of tasks across varied domains and models and is a pioneering effort in applying interpretability for scalable internal oversight of LMs.