Linguistic Neuron Overlap Patterns to Facilitate Cross-lingual Transfer on Low-resource Languages

The current Large Language Models (LLMs) face significant challenges in improving performance on low-resource languages and urgently need data-efficient methods without costly fine-tuning. From the perspective of language-bridge, we propose BridgeX-ICL, a simple yet effective method to improve zero-shot Cross-lingual In-Context Learning (X-ICL) for low-resource languages. Unlike existing works focusing on language-specific neurons, BridgeX-ICL explores whether sharing neurons can improve cross-lingual performance in LLMs or not. We construct neuron probe data from the ground-truth MUSE bilingual dictionaries, and define a subset of language overlap neurons accordingly, to ensure full activation of these anchored neurons. Subsequently, we propose an HSIC-based metric to quantify LLMs' internal linguistic spectrum based on overlap neurons, which guides optimal bridge selection. The experiments conducted on 2 cross-lingual tasks and 15 language pairs from 7 diverse families (covering both high-low and moderate-low pairs) validate the effectiveness of BridgeX-ICL and offer empirical insights into the underlying multilingual mechanisms of LLMs.

Following the Whispers of Values: Unraveling Neural Mechanisms Behind Value-Oriented Behaviors in LLMs

Despite the impressive performance of large language models (LLMs), they can present unintended biases and harmful behaviors driven by encoded values, emphasizing the urgent need to understand the value mechanisms behind them. However, current research primarily evaluates these values through external responses with a focus on AI safety, lacking interpretability and failing to assess social values in real-world contexts. In this paper, we propose a novel framework called ValueExploration, which aims to explore the behavior-driven mechanisms of National Social Values within LLMs at the neuron level. As a case study, we focus on Chinese Social Values and first construct C-voice, a large-scale bilingual benchmark for identifying and evaluating Chinese Social Values in LLMs. By leveraging C-voice, we then identify and locate the neurons responsible for encoding these values according to activation difference. Finally, by deactivating these neurons, we analyze shifts in model behavior, uncovering the internal mechanism by which values influence LLM decision-making. Extensive experiments on four representative LLMs validate the efficacy of our framework. The benchmark and code will be available.

A Survey on Multilingual Large Language Models: Corpora, Alignment, and Bias

Based on the foundation of Large Language Models (LLMs), Multilingual Large Language Models (MLLMs) have been developed to address the challenges of multilingual natural language processing tasks, hoping to achieve knowledge transfer from high-resource to low-resource languages. However, significant limitations and challenges still exist, such as language imbalance, multilingual alignment, and inherent bias. In this paper, we aim to provide a comprehensive analysis of MLLMs, delving deeply into discussions surrounding these critical issues. First of all, we start by presenting an overview of MLLMs, covering their evolution, key techniques, and multilingual capacities. Secondly, we explore widely utilized multilingual corpora for MLLMs' training and multilingual datasets oriented for downstream tasks that are crucial for enhancing the cross-lingual capability of MLLMs. Thirdly, we survey the existing studies on multilingual representations and investigate whether the current MLLMs can learn a universal language representation. Fourthly, we discuss bias on MLLMs including its category and evaluation metrics, and summarize the existing debiasing techniques. Finally, we discuss existing challenges and point out promising research directions. By demonstrating these aspects, this paper aims to facilitate a deeper understanding of MLLMs and their potentiality in various domains.

Deep MSFOP: Multiple Spectral filter Operators Preservation in Deep Functional Maps for Unsupervised Shape Matching

We propose a novel constraint called Multiple Spectral filter Operators Preservation (MSFOR) to compute functional maps and based on it, develop an efficient deep functional map architecture called Deep MSFOP for shape matching. The core idea is that, instead of using the general descriptor preservation constraint, we require our maps to preserve multiple spectral filter operators. This allows us to incorporate more informative geometrical information, contained in different frequency bands of functions, into the functional map computing. This can be confirmed by that some previous techniques like wavelet preservation and LBO commutativity are actually our special cases. Moreover, we also develop a very efficient way to compute the maps with MSFOP constraint, which can be conveniently embedded into the deep learning, especially having learnable filter operators. Utilizing the above results, we finally design our Deep MSFOP pipeline, equipped with a suitable unsupervised loss jointly penalizing the functional map and the underlying pointwise map. Our deep functional map has notable advantages, including that the functional map is more geometrically informative and guaranteed to be proper, and the computing is numerically stable. Extensive experimental results on different datasets demonstrate that our approach outperforms the existing state-of-the-art methods, especially in challenging settings like non-isometric and inconsistent topology datasets.