arXiv subjects

Shichang Zhang

Publications and source records attributed to Shichang Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Position: Explainability Research Must Prioritize Foundations over Ad-hoc Methods

Despite the proliferation of Explainable AI (XAI) techniques -- from feature attributions to sparse autoencoders -- explanations rarely influence real-world workflows. In practice, they are often generated and discarded without guiding meaningful action. This gap reflects foundational shortcomings: research has not yet established methodologies for integrating explanations into end-to-end, human-in-the-loop systems. This position paper argues that the machine learning community must pivot from ad-hoc XAI methods toward addressing foundational & structural challenges, including unclear problem formulations, underspecified evaluation objectives, and the absence of pipelines for explanation-driven feedback. We support this claim through an analysis of recent ICML, NeurIPS, and ICLR papers and a survey of XAI practitioners, revealing recurring issues that limit cumulative progress. We conclude by outlining a practical checklist designed to shift XAI toward a more human-centered, action-oriented paradigm. By emphasizing foundational clarity over the development of ad-hoc methods, we hope to provide a roadmap for integrating explanations into actionable, feedback-driven AI systems.

cs.LG↗

arXiv2026

How Faithful Is Trajectory-Based Data Attribution? Error Sources, Remedies, and Practical Guidelines

Trajectory-based data attribution methods estimate the influence of training samples on model predictions by unrolling the training trajectory. They are widely used in applications such as data selection, data valuation, and model diagnosis, but there is a lack of comprehensive error analysis of these methods, raising concerns about method faithfulness and hindering reliable deployment. In this work, we provide the first systematic analysis of error sources in trajectory-based data attribution, together with concrete remedies to mitigate them and practical guidelines for downstream use. We organize the total error into three categories, config-level, algorithm-level, and system-level. We make three contributions. First, we identify optimizer mismatch as the dominant config-level error: existing methods derive their attribution under the assumption of SGD, even for models trained with the modern de facto optimizer AdamW. We propose AdamW-influence to fully account for AdamW's optimization dynamics, yielding improvements from 10% to over 300% in Spearman correlation between estimated and ground-truth influence across four settings spanning MLP, CNN, GPT-2, and Llama 3.2-1B. Second, we isolate the remaining algorithm-level error arising from the first-order Taylor approximation, identify the learning rate and trajectory length as factors governing the error magnitude, and derive a closed-form error proxy that can be evaluated along the original trajectory without retraining. Third, we translate these insights into practical guidelines for data selection by unifying offline and online strategies under a K-step look-ahead framework. Under this framework, online selection with a short horizon often matches or exceeds offline, and the optimal horizon can be tuned jointly with the learning rate. Together, these results turn the framework into an actionable selection recipe for practitioners.

cs.LG↗

arXiv2025

From Indirect Object Identification to Syllogisms: Exploring Binary Mechanisms in Transformer Circuits

Transformer-based language models (LMs) can perform a wide range of tasks, and mechanistic interpretability (MI) aims to reverse engineer the components responsible for task completion to understand their behavior. Previous MI research has focused on linguistic tasks such as Indirect Object Identification (IOI). In this paper, we investigate the ability of GPT-2 small to handle binary truth values by analyzing its behavior with syllogistic prompts, e.g., "Statement A is true. Statement B matches statement A. Statement B is", which requires more complex logical reasoning compared to IOI. Through our analysis of several syllogism tasks of varying difficulty, we identify multiple circuits that mechanistically explain GPT-2's logical-reasoning capabilities and uncover binary mechanisms that facilitate task completion, including the ability to produce a negated token not present in the input prompt through negative heads. Our evaluation using a faithfulness metric shows that a circuit comprising five attention heads achieves over 90% of the original model's performance. By relating our findings to IOI analysis, we provide new insights into the roles of specific attention heads and MLPs in LMs. These insights contribute to a broader understanding of model reasoning and support future research in mechanistic interpretability.

cs.CL↗

arXiv2025

Retention analysis of edited knowledge after fine-tuning

Large language models (LLMs) store vast amounts of knowledge, which often requires updates to correct factual errors, incorporate newly acquired information, or adapt model behavior. Model editing methods have emerged as efficient solutions for such updates, offering localized and precise knowledge modification at significantly lower computational cost than continual training. In parallel, LLMs are frequently fine-tuned for a wide range of downstream tasks. However, the effect of fine-tuning on previously edited knowledge remains poorly understood. In this work, we systematically investigate how different fine-tuning objectives interact with various model editing techniques. Our findings show that edited knowledge is substantially more susceptible to forgetting during fine-tuning than intrinsic knowledge acquired through pre-training. This analysis highlights a key limitation of current editing approaches and suggests that evaluating edit robustness under downstream fine-tuning is critical for their practical deployment. We further find that knowledge retention can be significantly improved by either augmenting edit knowledge with paraphrases or by freezing layers associated with edited content in fine-tuning stage, offering insight for developing more robust editing algorithms.

cs.CL↗

arXiv2025

FUSE: Measure-Theoretic Compact Fuzzy Set Representation for Taxonomy Expansion

Taxonomy Expansion, which models complex concepts and their relations, can be formulated as a set representation learning task. The generalization of set, fuzzy set, incorporates uncertainty and measures the information within a semantic concept, making it suitable for concept modeling. Existing works usually model sets as vectors or geometric objects such as boxes, which are not closed under set operations. In this work, we propose a sound and efficient formulation of set representation learning based on its volume approximation as a fuzzy set. The resulting embedding framework, Fuzzy Set Embedding (FUSE), satisfies all set operations and compactly approximates the underlying fuzzy set, hence preserving information while being efficient to learn, relying on minimum neural architecture. We empirically demonstrate the power of FUSE on the task of taxonomy expansion, where FUSE achieves remarkable improvements up to 23% compared with existing baselines. Our work marks the first attempt to understand and efficiently compute the embeddings of fuzzy sets.

cs.LG↗

arXiv2025

Who Gets Credit or Blame? Attributing Accountability in Modern AI Systems

Modern AI systems are typically developed through multiple stages-pretraining, fine-tuning rounds, and subsequent adaptation or alignment, where each stage builds on the previous ones and updates the model in distinct ways. This raises a critical question of accountability: when a deployed model succeeds or fails, which stage is responsible, and to what extent? We pose the accountability attribution problem for tracing model behavior back to specific stages of the model development process. To address this challenge, we propose a general framework that answers counterfactual questions about stage effects: how would the model's behavior have changed if the updates from a particular stage had not occurred? Within this framework, we introduce estimators that efficiently quantify stage effects without retraining the model, accounting for both the data and key aspects of model optimization dynamics, including learning rate schedules, momentum, and weight decay. We demonstrate that our approach successfully quantifies the accountability of each stage to the model's behavior. Based on the attribution results, our method can identify and remove spurious correlations learned during image classification and text toxicity detection tasks that were developed across multiple stages. Our approach provides a practical tool for model analysis and represents a significant step toward more accountable AI development.

cs.LG↗

arXiv2025

Heuristic Methods are Good Teachers to Distill MLPs for Graph Link Prediction

Link prediction is a crucial graph-learning task with applications including citation prediction and product recommendation. Distilling Graph Neural Networks (GNNs) teachers into Multi-Layer Perceptrons (MLPs) students has emerged as an effective approach to achieve strong performance and reducing computational cost by removing graph dependency. However, existing distillation methods only use standard GNNs and overlook alternative teachers such as specialized model for link prediction (GNN4LP) and heuristic methods (e.g., common neighbors). This paper first explores the impact of different teachers in GNN-to-MLP distillation. Surprisingly, we find that stronger teachers do not always produce stronger students: MLPs distilled from GNN4LP can underperform those distilled from simpler GNNs, while weaker heuristic methods can teach MLPs to near-GNN performance with drastically reduced training costs. Building on these insights, we propose Ensemble Heuristic-Distilled MLPs (EHDM), which eliminates graph dependencies while effectively integrating complementary signals via a gating mechanism. Experiments on ten datasets show an average 7.93% improvement over previous GNN-to-MLP approaches with 1.95-3.32 times less training time, indicating EHDM is an efficient and effective link prediction method. Our code is available at https://github.com/ZongyueQin/EHDM

cs.LG↗

arXiv2025

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence

Post-training is essential for the success of large language models (LLMs), transforming pre-trained base models into more useful and aligned post-trained models. While plenty of works have studied post-training algorithms and evaluated post-training models by their outputs, it remains understudied how post-training reshapes LLMs internally. In this paper, we compare base and post-trained LLMs mechanistically from four perspectives to better understand post-training effects. Our findings across model families and datasets reveal that: (1) Post-training does not change the factual knowledge storage locations, and it adapts knowledge representations from the base model while developing new knowledge representations; (2) Both truthfulness and refusal can be represented by vectors in the hidden representation space. The truthfulness direction is highly similar between the base and post-trained model, and it is effectively transferable for interventions; (3) The refusal direction is different between the base and post-trained models, and it shows limited forward transferability; (4) Differences in confidence between the base and post-trained models cannot be attributed to entropy neurons. Our study provides insights into the fundamental mechanisms preserved and altered during post-training, facilitates downstream tasks like model steering, and could potentially benefit future research in interpretability and LLM post-training. Our code is publicly available at https://github.com/HZD01/post-training-mechanistic-analysis.

cs.CL↗

arXiv2025

Large Language Model Agent: A Survey on Methodology, Applications and Challenges

The era of intelligent agents is upon us, driven by revolutionary advancements in large language models. Large Language Model (LLM) agents, with goal-driven behaviors and dynamic adaptation capabilities, potentially represent a critical pathway toward artificial general intelligence. This survey systematically deconstructs LLM agent systems through a methodology-centered taxonomy, linking architectural foundations, collaboration mechanisms, and evolutionary pathways. We unify fragmented research threads by revealing fundamental connections between agent design principles and their emergent behaviors in complex environments. Our work provides a unified architectural perspective, examining how agents are constructed, how they collaborate, and how they evolve over time, while also addressing evaluation methodologies, tool applications, practical challenges, and diverse application domains. By surveying the latest developments in this rapidly evolving field, we offer researchers a structured taxonomy for understanding LLM agents and identify promising directions for future research. The collection is available at https://github.com/luo-junyu/Awesome-Agent-Papers.

cs.CL↗

arXiv2025

Towards Unified Attribution in Explainable AI, Data-Centric AI, and Mechanistic Interpretability

The increasing complexity of AI systems has made understanding their behavior critical. Numerous interpretability methods have been developed to attribute model behavior to three key aspects: input features, training data, and internal model components, which emerged from explainable AI, data-centric AI, and mechanistic interpretability, respectively. However, these attribution methods are studied and applied rather independently, resulting in a fragmented landscape of methods and terminology. This position paper argues that feature, data, and component attribution methods share fundamental similarities, and a unified view of them benefits both interpretability and broader AI research. To this end, we first analyze popular methods for these three types of attributions and present a unified view demonstrating that these seemingly distinct methods employ similar techniques (such as perturbations, gradients, and linear approximations) over different aspects and thus differ primarily in their perspectives rather than techniques. Then, we demonstrate how this unified view enhances understanding of existing attribution methods, highlights shared concepts and evaluation criteria among these methods, and leads to new research directions both in interpretability research, by addressing common challenges and facilitating cross-attribution innovation, and in AI more broadly, with applications in model editing, steering, and regulation.

cs.LG↗

arXiv2024

Generalized Group Data Attribution

Data Attribution (DA) methods quantify the influence of individual training data points on model outputs and have broad applications such as explainability, data selection, and noisy label identification. However, existing DA methods are often computationally intensive, limiting their applicability to large-scale machine learning models. To address this challenge, we introduce the Generalized Group Data Attribution (GGDA) framework, which computationally simplifies DA by attributing to groups of training points instead of individual ones. GGDA is a general framework that subsumes existing attribution methods and can be applied to new DA techniques as they emerge. It allows users to optimize the trade-off between efficiency and fidelity based on their needs. Our empirical results demonstrate that GGDA applied to popular DA methods such as Influence Functions, TracIn, and TRAK results in upto 10x-50x speedups over standard DA methods while gracefully trading off attribution fidelity. For downstream applications such as dataset pruning and noisy label identification, we demonstrate that GGDA significantly improves computational efficiency and maintains effectiveness, enabling practical applications in large-scale machine learning scenarios that were previously infeasible.

cs.LG↗

arXiv2024

Predicting and Interpreting Energy Barriers of Metallic Glasses with Graph Neural Networks

Metallic Glasses (MGs) are widely used materials that are stronger than steel while being shapeable as plastic. While understanding the structure-property relationship of MGs remains a challenge in materials science, studying their energy barriers (EBs) as an intermediary step shows promise. In this work, we utilize Graph Neural Networks (GNNs) to model MGs and study EBs. We contribute a new dataset for EB prediction and a novel Symmetrized GNN (SymGNN) model that is E(3)-invariant in expectation. SymGNN handles invariance by aggregating over orthogonal transformations of the graph structure. When applied to EB prediction, SymGNN are more accurate than molecular dynamics (MD) local-sampling methods and other machine-learning models. Compared to precise MD simulations, SymGNN reduces the inference time on new MGs from roughly 41 days to less than one second. We apply explanation algorithms to reveal the relationship between structures and EBs. The structures that we identify through explanations match the medium-range order (MRO) hypothesis and possess unique topological properties. Our work enables effective prediction and interpretation of MG EBs, bolstering material science research.

cond-mat.dis-nn↗

arXiv2024

SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models

Most of the existing Large Language Model (LLM) benchmarks on scientific problem reasoning focus on problems grounded in high-school subjects and are confined to elementary algebraic operations. To systematically examine the reasoning capabilities required for solving complex scientific problems, we introduce an expansive benchmark suite SciBench for LLMs. SciBench contains a carefully curated dataset featuring a range of collegiate-level scientific problems from mathematics, chemistry, and physics domains. Based on the dataset, we conduct an in-depth benchmarking study of representative open-source and proprietary LLMs with various prompting strategies. The results reveal that the current LLMs fall short of delivering satisfactory performance, with the best overall score of merely 43.22%. Furthermore, through a detailed user study, we categorize the errors made by LLMs into ten problem-solving abilities. Our analysis indicates that no single prompting strategy significantly outperforms the others and some strategies that demonstrate improvements in certain problem-solving skills could result in declines in other skills. We envision that SciBench will catalyze further developments in the reasoning abilities of LLMs, thereby ultimately contributing to scientific research and discovery.

cs.CL↗

arXiv2024

Automated Molecular Concept Generation and Labeling with Large Language Models

Artificial intelligence (AI) is transforming scientific research, with explainable AI methods like concept-based models (CMs) showing promise for new discoveries. However, in molecular science, CMs are less common than black-box models like Graph Neural Networks (GNNs), due to their need for predefined concepts and manual labeling. This paper introduces the Automated Molecular Concept (AutoMolCo) framework, which leverages Large Language Models (LLMs) to automatically generate and label predictive molecular concepts. Through iterative concept refinement, AutoMolCo enables simple linear models to outperform GNNs and LLM in-context learning on several benchmarks. The framework operates without human knowledge input, overcoming limitations of existing CMs while maintaining explainability and allowing easy intervention. Experiments on MoleculeNet and High-Throughput Experimentation (HTE) datasets demonstrate that AutoMolCo-induced explainable CMs are beneficial for molecular science research.

cs.AI↗

arXiv2024

Hierarchical Compression of Text-Rich Graphs via Large Language Models

Text-rich graphs, prevalent in data mining contexts like e-commerce and academic graphs, consist of nodes with textual features linked by various relations. Traditional graph machine learning models, such as Graph Neural Networks (GNNs), excel in encoding the graph structural information, but have limited capability in handling rich text on graph nodes. Large Language Models (LLMs), noted for their superior text understanding abilities, offer a solution for processing the text in graphs but face integration challenges due to their limitation for encoding graph structures and their computational complexities when dealing with extensive text in large neighborhoods of interconnected nodes. This paper introduces ``Hierarchical Compression'' (HiCom), a novel method to align the capabilities of LLMs with the structure of text-rich graphs. HiCom processes text in a node's neighborhood in a structured manner by organizing the extensive textual information into a more manageable hierarchy and compressing node text step by step. Therefore, HiCom not only preserves the contextual richness of the text but also addresses the computational challenges of LLMs, which presents an advancement in integrating the text processing power of LLMs with the structural complexities of text-rich graphs. Empirical results show that HiCom can outperform both GNNs and LLM backbones for node classification on e-commerce and citation graphs. HiCom is especially effective for nodes from a dense region in a graph, where it achieves a 3.48% average performance improvement on five datasets while being more efficient than LLM backbones.

cs.SI↗

arXiv2024

Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller

We propose SelfControl, an inference-time model control method utilizing gradients to control the behavior of large language models (LLMs) without explicit human annotations. Given a desired behavior expressed in a natural language suffix string concatenated to the input prompt, SelfControl computes gradients of the LLM's self-evaluation of the suffix with respect to its latent representations. The gradients are used to directly control the auto-regressive generation process towards desired behaviors, which eliminates human supervision, achieves precise and transparent control, and offers on-the-fly adaptability. To further enhance efficiency, we introduce SelfControl_{Prefix}, a compact module that encapsulates the learned representations from gradients into a SelfControl_{Prefix}, facilitating efficient inference-time control with no latency compared to the original model and allowing control for multiple behaviors simultaneously. Our experiments demonstrate SelfControl's efficacy across multiple domains, where it improves over SOTA for 8.3% in detoxification, 3.1% in truthfulness enhancement, 4%~10% in controlling on emotion tones, and 48.2% in privacy protection, i.e., completely remove privacy leakage issue. Additionally, we demonstrate that SelfControl can be used for data synthesis and to improve reasoning abilities.

cs.CL↗

arXiv2024

Efficient Ensembles Improve Training Data Attribution

Training data attribution (TDA) methods aim to quantify the influence of individual training data points on the model predictions, with broad applications in data-centric AI, such as mislabel detection, data selection, and copyright compensation. However, existing methods in this field, which can be categorized as retraining-based and gradient-based, have struggled with the trade-off between computational efficiency and attribution efficacy. Retraining-based methods can accurately attribute complex non-convex models but are computationally prohibitive, while gradient-based methods are efficient but often fail for non-convex models. Recent research has shown that augmenting gradient-based methods with ensembles of multiple independently trained models can achieve significantly better attribution efficacy. However, this approach remains impractical for very large-scale applications. In this work, we discover that expensive, fully independent training is unnecessary for ensembling the gradient-based methods, and we propose two efficient ensemble strategies, DROPOUT ENSEMBLE and LORA ENSEMBLE, alternative to naive independent ensemble. These strategies significantly reduce training time (up to 80%), serving time (up to 60%), and space cost (up to 80%) while maintaining similar attribution efficacy to the naive independent ensemble. Our extensive experimental results demonstrate that the proposed strategies are effective across multiple TDA methods on diverse datasets and models, including generative settings, significantly advancing the Pareto frontier of TDA methods with better computational efficiency and attribution efficacy.

cs.LG↗

arXiv2024

Parameter-Efficient Tuning Large Language Models for Graph Representation Learning

Text-rich graphs, which exhibit rich textual information on nodes and edges, are prevalent across a wide range of real-world business applications. Large Language Models (LLMs) have demonstrated remarkable abilities in understanding text, which also introduced the potential for more expressive modeling in text-rich graphs. Despite these capabilities, efficiently applying LLMs to representation learning on graphs presents significant challenges. Recently, parameter-efficient fine-tuning methods for LLMs have enabled efficient new task generalization with minimal time and memory consumption. Inspired by this, we introduce Graph-aware Parameter-Efficient Fine-Tuning - GPEFT, a novel approach for efficient graph representation learning with LLMs on text-rich graphs. Specifically, we utilize a graph neural network (GNN) to encode structural information from neighboring nodes into a graph prompt. This prompt is then inserted at the beginning of the text sequence. To improve the quality of graph prompts, we pre-trained the GNN to assist the frozen LLM in predicting the next token in the node text. Compared with existing joint GNN and LMs, our method directly generate the node embeddings from large language models with an affordable fine-tuning cost. We validate our approach through comprehensive experiments conducted on 8 different text-rich graphs, observing an average improvement of 2% in hit@1 and Mean Reciprocal Rank (MRR) in link prediction evaluations. Our results demonstrate the efficacy and efficiency of our model, showing that it can be smoothly integrated with various large language models, including OPT, LLaMA and Falcon.

cs.CL↗