George Ma, Anurag Koul, Qi Chen, Yawen Wu, Sachit Kuhar, Yu Yu, Aritra Sengupta, Varun Kumar, Murali Krishna Ramanathan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Mechanistic interpretability research seeks to reverse-engineer large language models (LLMs) by uncovering the internal representations of concepts within their activations. Sparse Autoencoders (SAEs) have emerged as a valuable tool for disentangling polysemantic neurons into more monosemantic, interpretable features. However, recent work on automatic explanation generation for these features has faced challenges: explanations tend to be overly broad and fail to take polysemanticity into consideration. This work addresses these limitations by introducing a similarity-based strategy for sourcing close negative sentences that more effectively falsify generated explanations. Additionally, we propose a structured, component-based format for feature explanations and a tree-based, iterative explanation method that refines explanations. We demonstrate that our structured format and tree-based explainer improve explanation quality, while our similarity-based evaluation strategy exposes biases in existing interpretability methods. We also analyze the evolution of feature complexity and polysemanticity across LLM layers, offering new insights into information content within LLMs' residual streams.

In many applications, we desire neural networks to exhibit invariance or equivariance to certain groups due to symmetries inherent in the data. Recently, frame-averaging methods emerged to be a unified framework for attaining symmetries efficiently by averaging over input-dependent subsets of the group, i.e., frames. What we currently lack is a principled understanding of the design of frames. In this work, we introduce a canonicalization perspective that provides an essential and complete view of the design of frames. Canonicalization is a classic approach for attaining invariance by mapping inputs to their canonical forms. We show that there exists an inherent connection between frames and canonical forms. Leveraging this connection, we can efficiently compare the complexity of frames as well as determine the optimality of certain frames. Guided by this principle, we design novel frames for eigenvectors that are strictly superior to existing methods --- some are even optimal --- both theoretically and empirically. The reduction to the canonicalization perspective further uncovers equivalences between previous methods. These observations suggest that canonicalization provides a fundamental understanding of existing frame-averaging methods and unifies existing equivariant and invariant learning methods. Code is available at https://github.com/PKU-ML/canonicalization.

Spectral embedding is a powerful graph embedding technique that has received a lot of attention recently due to its effectiveness on Graph Transformers. However, from a theoretical perspective, the universal expressive power of spectral embedding comes at the price of losing two important invariance properties of graphs, sign and basis invariance, which also limits its effectiveness on graph data. To remedy this issue, many previous methods developed costly approaches to learn new invariants and suffer from high computation complexity. In this work, we explore a minimal approach that resolves the ambiguity issues by directly finding canonical directions for the eigenvectors, named Laplacian Canonization (LC). As a pure pre-processing method, LC is light-weighted and can be applied to any existing GNNs. We provide a thorough investigation, from theory to algorithm, on this approach, and discover an efficient algorithm named Maximal Axis Projection (MAP) that works for both sign and basis invariance and successfully canonizes more than 90\% of all eigenvectors. Experiments on real-world benchmark datasets like ZINC, MOLTOX21, and MOLPCBA show that MAP consistently outperforms existing methods while bringing minimal computation overhead. Code is available at https://github.com/PKU-ML/LaplacianCanonization.

Corporations spend billions of dollars annually caring for customers across multiple contact channels. A customer journey is the complete sequence of contacts that a given customer has with a company across multiple channels of communication. While each contact is important and contains rich information, studying customer journeys provides a better context to understand customers' behavior in order to improve customer satisfaction and loyalty, and to reduce care costs. However, journey sequences have a complex format due to the heterogeneity of user behavior: they are variable-length, multi-attribute, and exhibit a large cardinality in categories (e.g. contact reasons). The question of how to characterize and learn representations of customer journeys has not been studied in the literature. We propose to learn journey embeddings using a sequence-to-sequence framework that converts each customer journey into a fixed-length latent embedding. In order to improve the disentanglement and distributional properties of embeddings, the model is further modified by incorporating a Wasserstein autoencoder inspired regularization on the distribution of embeddings. Experiments conducted on an enterprise-scale dataset demonstrate the effectiveness of the proposed model and reveal significant improvements due to the regularization in both distinguishing journey pattern characteristics and predicting future customer engagement.