Backward compatible representation learning enables updated models to integrate seamlessly with existing ones, avoiding to reprocess stored data. Despite recent advances, existing compatibility approaches in Euclidean space neglect the uncertainty in the old embedding models and force the new model to replicate outdated representations regardless of their quality, and thereby hindering the learning process. In this paper, we switch perspectives to hyperbolic geometry, where we treat time as a natural axis for capturing a model’s confidence and evolution. By lifting embeddings into hyperbolic space and constraining updated embeddings to lie within the entailment cone of the old ones, we maintain generational consistency across models while accounting for uncertainties in the representations. To further enhance compatibility, we introduce a robust contrastive alignment loss that dynamically adjusts alignment weights based on the uncertainty of the old embeddings. Experiments validate the superiority of the proposed method in achieving compatibility, paving the way for more resilient and adaptable machine learning systems.

Large Language Models (LLMs) matrices can often be expressed in low-rank format with potential to relax memory and compute resource requirements. Unlike previous works which pivot around developing novel matrix decomposition algorithms, in this work we focus to study the emerging non-uniform low-rank properties across weight matrices in LLMs through the lens of stabilizing gradient subspace. \textit{Firstly,} we provide a theoretical framework to understand the stabilization of gradient subspaces through Hessian analysis. \textit{Secondly,} we empirically establish a consequential relationship between the gradient dynamics and low-rank expressiveness of weight matrices. Our findings reveal that different LLM components exhibit varying levels of converged low-rank structure, necessitating a non-uniform rank reduction across them to minimize performance drop due to compression. In view of that, we present \textit{Weight Low-Rank Projection} \textbf{(WeLore)} that unifies weight compression and memory-efficient fine-tuning as ONE, in a data-agnostic and one-shot way. Going beyond only as a compression technique, WeLore categorizes weight matrices into Low-rank Components (LRCs) and Non-Low-rank Components (N-LRCs) based on their ability to express themselves as low-rank. Our gradient dynamics perspective illustrate that \textit{LRCs tend to have better finetuning capabilities} and their standalone finetuning can closely mimic (sometimes outperform) the training loss trajectory and performance of full-finetuning with notable memory and compute footprint reduction. All codes and checkpoints will be released.

Runjin Chen, Gabriel Jacob Perin, Xuxi Chen, Xilun Chen, Yan Han, Nina S. T. Hirata, Junyuan Hong, Bhavya Kailkhura. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Deep recommender systems rely heavily on large embedding tables to handle high-cardinality categorical features such as user/item identifiers, and face significant memory constraints at scale.To tackle this challenge, hashing techniques are often employed to map multiple entities to the same embedding and thus reduce the size of the embedding tables.Concurrently, graph-based collaborative signals have emerged as powerful tools in recommender systems, yet their potential for optimizing embedding table reduction remains unexplored.This paper introduces GraphHash, the first graph-based approach that leverages modularity-based bipartite graph clustering on user-item interaction graphs to reduce embedding table sizes.We demonstrate that the modularity objective has a theoretical connection to message-passing, which provides a foundation for our method.By employing fast clustering algorithms, GraphHash serves as a computationally efficient proxy for message-passing during preprocessing and a plug-andplay graph-based alternative to traditional ID hashing.Extensive experiments show that GraphHash substantially outperforms diverse hashing baselines on both retrieval and click-through-rate prediction tasks.In particular, GraphHash achieves on average a 101.52% improvement in recall when reducing the embedding table size by more than 75%, highlighting the value of graph-based collaborative information for model reduction.

Graph Neural Networks (GNNs) have empowered the advance in graph-structured data analysis. Recently, the rise of Large Language Models (LLMs) like GPT-4 has heralded a new era in deep learning. However, their application to graph data poses distinct challenges due to the inherent difficulty of translating graph structures to language. To this end, we introduce the the **L**arge **L**anguage **a**nd **G**raph **A**ssistant (**LLaGA**), an innovative model that effectively integrates LLM capabilities to handle the complexities of graph-structured data. LLaGA retains the general-purpose nature of LLMs while adapting graph data into a format compatible with LLM input. LLaGA achieves this by reorganizing graph nodes to structure-aware sequences and then mapping these into the token embedding space through a versatile projector. LLaGA excels in versatility, generalizability and interpretability, allowing it to perform consistently well across different datasets and tasks, extend its ability to unseen datasets or tasks, and provide explanations for graphs. Our extensive experiments across popular graph benchmarks show that LLaGA delivers outstanding performance across four datasets and three tasks using one single model, surpassing state-of-the-art graph models in both supervised and zero-shot scenarios.

This paper aims to overcome the ``lost-in-the-middle'' challenge of large language models (LLMs). While recent advancements have successfully enabled LLMs to perform stable language modeling with up to 4 million tokens, the persistent difficulty faced by most LLMs in identifying relevant information situated in the middle of the context has not been adequately tackled. To address this problem, this paper introduces Multi-scale Positional Encoding (Ms-PoE) which is a simple yet effective plug-and-play approach to enhance the capacity of LLMs to handle the relevant information located in the middle of the context, without fine-tuning or introducing any additional overhead. Ms-PoE leverages the position indice rescaling to relieve the long-term decay effect introduced by RoPE, while meticulously assigning distinct scaling ratios to different attention heads to preserve essential knowledge learned during the pre-training step, forming a multi-scale context fusion from short to long distance. Extensive experiments with a wide range of LLMs demonstrate the efficacy of our approach. Notably, Ms-PoE achieves an average accuracy gain of up to 3.8 on the Zero-SCROLLS benchmark over the original LLMs. Code will be made public upon acceptence.

Relational databases are the main storage model of structured data in most businesses, which usually involves multiple tables with key-foreign-key relationships. In practice, data analysts often want to pose predictive classification queries over relational databases. To answer such queries, many existing approaches perform supervised learning to train classification models, which heavily rely on the availability of sufficient labeled data. In this paper, we propose a novel graph-based contrastive framework for semi-supervised learning on relational databases, achieving promising predictive classification performance with only a handful of labeled data. Our framework utilizes contrastive learning to exploit additional supervision signals from massive unlabeled data. Specifically, we develop two contrastive graph views that are 1) advantageous for modeling complex relationships and correlations among structured data in a relational database, and 2) complementary to each other for learning robust representations of structured data to be classified. We also leverage label information in contrastive learning to mitigate its negative effect in knowledge transfer on the supervised counterpart. We conduct extensive experiments on three real-world relational databases and the results demonstrate that our framework is able to achieve the state-of-the-art predictive performance in limited labeled data settings, compared with various supervised and semi-supervised learning approaches.

Entity Matching is a classic research problem in any data analytics pipeline, aiming to identify records referring to the same real-world entity. It plays an important role in data cleansing and integration. Advanced entity matching techniques focus on extracting syntactic or semantic features from record pairs via complex neural architectures or pre-trained language models. However, the performances always suffer from noisy or missing attribute values in the records. We observe that comparing one record with several relevant records in a collective manner allows each pairwise matching decision to be made by borrowing valuable insights from other pairs, which is beneficial to the overall matching performance. In this paper, we propose a generic one-to-set neural framework named GNEM for entity matching. GNEM predicts matching labels between one record and a set of relevant records simultaneously. It constructs a record pair graph with weighted edges and adopts the graph neural network to propagate information among pairs. We further show that GNEM can be interpreted as an extension and generalization of the existing pairwise matching techniques. Extensive experiments on real-world data sets demonstrate that GNEM consistently outperforms the existing pairwise entity matching techniques and achieves up to 8.4% improvement on F1-Score compared with the state-of-the-art neural methods.

This paper presents a method to pursue a semantic and quantitative explanation for the knowledge encoded in a convolutional neural network (CNN). The estimation of the specific rationale of each prediction made by the CNN presents a key issue of understanding neural networks, and it is of significant values in real applications. In this study, we propose to distill knowledge from the CNN into an explainable additive model, which explains the CNN prediction quantitatively. We discuss the problem of the biased interpretation of CNN predictions. To overcome the biased interpretation, we develop prior losses to guide the learning of the explainable additive model. Experimental results have demonstrated the effectiveness of our method.

We define a unified information-based measure to provide quantitative explanations on how intermediate layers of deep Natural Language Processing (NLP) models leverage information of input words. Our method advances existing explanation methods by addressing issues in coherency and generality. Explanations generated by using our method are consistent and faithful across different timestamps, layers, and models. We show how our method can be applied to four widely used models in NLP and explain their performances on three real-world benchmark datasets.