Singapore

Fine-tuning large language models (LLMs) in a parameter-efficient manner while preserving their pre-trained world knowledge remains a significant challenge. While Low-Rank Adaptation (LoRA) and its variants effectively mitigate catastrophic forgetting, they do not fully eliminate the loss of critical pre-trained knowledge. In this work, we first analyze the layer-wise distribution of domain-specific knowledge within LLMs through knowledge localization, and empirically identify a clear layer-specific pattern: pre-trained world knowledge predominantly resides in lower layers, whereas knowledge relevant to downstream tasks is more concentrated in higher layers. Motivated by this observation, we propose L2-LoRA, a simple yet effective variant of LoRA that applies layer-specific $L_2$ regularization to the LoRA weights during fine-tuning. Specifically, L2-LoRA imposes stronger regularization on lower layers to preserve pre-trained world knowledge, while allowing greater adaptation in higher layers to better align with downstream tasks. Experiments across multiple benchmarks show that L2-LoRA not only consistently outperforms vanilla LoRA in downstream performance, but also effectively mitigates catastrophic forgetting by retaining more pre-trained knowledge.

AAAI 2026

L2-LoRA: Improving Low-Rank Adaptation with Layer-Specific Regularization

learning & optimization

large language models

interpretability

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

poster

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Heterogeneous graphs are widely used to model real-world systems with diverse entity types and relational structures, and existing methods have shown promising performance in various applications. However, most current models assume balanced and semantically aligned features across nodes, which rarely holds in practice. In scenarios such as social risk governance, node types often exhibit severe feature imbalance, making it difficult for standard aggregation mechanisms to extract meaningful signals. This imbalance leads to three key challenges: inaccurate neighbor weighting, noise propagation, and biased representations skewed toward text-rich nodes. To address these issues, we propose HeCoGNN, a collaborative and adaptive aggregation framework that jointly performs neighbor filtering and relation-aware message calibration, enabling robust representation learning under semantic disparity. Experiments on real-world social governance graphs show that HeCoGNN consistently outperforms state-of-the-art baselines, particularly in handling underrepresented and noisy node types.

Mitigating Noise and Imbalance in Social Governance Graphs for Multi-Type Risk Assessment

Multi-domain machine translation (MDMT) aims to build a unified model capable of translating content across diverse domains. Despite the impressive machine translation capabilities demonstrated by large language models (LLMs), domain adaptation still remains a challenge for LLMs. Existing MDMT methods such as in-context learning and parameter-efficient fine-tuning often suffer from domain shift, parameter interference and limited generalization. In this work, we propose a neuron-efficient fine-tuning framework for MDMT that identifies and updates consensus-aligned neurons within LLMs. These neurons are selected by maximizing the mutual information between neuron behavior and domain features, enabling LLMs to capture both generalizable translation patterns and domain-specific nuances. Our method then fine-tunes LLMs guided by these neurons, effectively mitigating parameter interference and domain-specific overfitting. Comprehensive experiments on three LLMs across ten German-English (De$\Rightarrow$En) and Chinese-English (Zh$\Rightarrow$En) translation domains evidence that our method consistently outperforms strong PEFT baselines on both seen and unseen domains, achieving state-of-the-art performance. The codes for this paper are available at https://anonymous.4open.science/r/CANEFT-3DAB.

Consensus-Aligned Neuron Efficient Fine-Tuning Large Language Models for Multi-Domain Machine Translation

Bayesian optimization (BO) has been widely used to optimize expensive and gradient-free objective functions across various domains. However, existing BO methods have not addressed the objective where both inputs and outputs are functions, which increasingly arise in complex systems as advanced sensing technologies. To fill this gap, we propose a novel function-on-function Bayesian optimization (FFBO) framework. Specifically, we first introduce a function-on-function Gaussian process (FFGP) model with an separable operator-valued kernel to capture the correlations between functional inputs and outputs. Compared to existing Gaussian process models, FFGP is modeled directly in the functional space. Based on FFGP, we define a scalar upper confidence bound (UCB) acquisition function using a weighted operator-based scalarization strategy. Then, a scalable functional gradient ascent algorithm (FGA) is developed to efficiently identify the optimal functional input. We further analyze the theoretical properties of the proposed method. Extensive experiments on synthetic and real-world data demonstrate the superior performance of FFBO over existing approaches.

Function-on-Function Bayesian Optimization

Deep learning has significantly advanced numerous fields by training on extensive annotated datasets. However, this data-driven paradigm faces limitations such as limited adaptability and high annotation costs, particularly when precise adherence to detailed, domain-specific guidelines is required in annotation. This challenge raises a critical question: Can models effectively shift from data-driven learning to autonomously leveraging guidelines with minimal annotated examples? To address this, we propose the Guideline-Driven Prompt (GDP) optimization framework, which shifts the learning paradigm from data-driven training to guideline-driven reasoning. GDP leverages Retrieval Augmented Generation (RAG) to retrieve essential fragments from complex guidelines and synthesize them into structured, executable prompts. A tree-based optimization algorithm systematically constructs and refines these prompts, explicitly capturing the intricate logic embedded in professional guidelines through a latent pipeline structure. Empirical evaluations on four datasets ranging from diverse domains and different tasks demonstrate that GDP effectively transitions the learning process from data-intensive methods to a guideline-driven approach in tasks requiring detailed and complex guideline adherence, reducing dependence on extensive annotated datasets.

Learning from Guidelines: Structured Prompt Optimization for Expert Annotation Tasks

Time series forecasting is essential in a wide range of real world applications. Recently, frequency-domain methods have attracted increasing interest for their ability to capture global dependencies. However, when applied to non-stationary time series, these methods encounter the $\textit{spectral entanglement}$ and the computational burden of complex-valued learning. The $\textit{spectral entanglement}$ refers to the overlap of trends, periodicities, and noise across the spectrum due to $\textit{spectral leakage}$ and the presence of non-stationarity. However, existing decompositions are not suited to resolving spectral entanglement. To address this, we propose the Frequency Decomposition Network (FreDN), which introduces a learnable Frequency Disentangler module to separate trend and periodic components directly in the frequency domain. Furthermore, we propose a theoretically supported ReIm Block to reduce the complexity of complex-valued operations while maintaining performance. We also re-examine the frequency-domain loss function and provide new theoretical insights into its effectiveness. Extensive experiments on seven long-term forecasting benchmarks demonstrate that FreDN outperforms state-of-the-art methods by up to 10\%. Furthermore, compared with standard complex-valued architectures, our real-imaginary shared-parameter design reduces the parameter count and computational cost by at least 50\%.

FreDN: Spectral Disentanglement for Time Series Forecasting via Learnable Frequency Decomposition

Contextual linear dueling bandits have recently garnered significant attention due to their widespread applications in important domains such as recommender systems and large language models. Classical dueling bandit algorithms are typically only applicable to a single agent. However, many applications of dueling bandits involve multiple agents who wish to collaborate for improved performance yet are unwilling to share their data. This motivates us to draw inspirations from $\textit{federated learning}$, which involves multiple agents aiming to collaboratively train their neural networks via gradient descent (GD) without sharing their raw data. Previous works have developed federated linear bandit algorithms which rely on closed-form updates of the bandit parameters (e.g., the linear function parameters) to achieve collaboration. However, in linear dueling bandits, the linear function parameters lack a closed-form expression and their estimation requires minimizing a loss function. This renders these previous methods inapplicable. In this work, we overcome this challenge through an innovative and principled combination of online gradient descent (OGD, for minimizing the loss function to estimate the linear function parameters) and federated learning, hence introducing our $\textit{federated linear dueling bandit with OGD} \texttt{(FLDB-OGD)}$ algorithm. Through rigorous theoretical analysis, we prove that $\textit{FLDB-OGD}$ enjoys a sub-linear upper bound on its cumulative regret and demonstrate a theoretical trade-off between regret and communication complexity. We conduct empirical experiments to demonstrate the effectiveness of $\textit{FLDB-OGD}$ and reveal valuable insights, such as the benefit of a larger number of agents, the regret-communication trade-off, among others.

Federated Linear Dueling Bandits

Machine unlearning, which selectively removes harmful knowledge from a pre-trained model without retraining from scratch, is crucial for addressing privacy, regulatory compliance, and ethical concerns in Large Language Models (LLMs). However, existing unlearning methods often struggle to thoroughly remove harmful knowledge, leaving residual harmful knowledge that can be easily recovered. To address these limitations, we propose Knowledge Density-Guided Unlearning via Blocks Reinsertion (KUnBR), a novel approach that first identifies layers with rich harmful knowledge and then thoroughly eliminates the harmful knowledge via re-insertion strategy. Our method introduces knowledge density estimation to quantify and locate layers containing the most harmful knowledge, enabling precise unlearning. Additionally, we design a layer re-insertion strategy that extracts and re-inserts harmful knowledge-rich layers into the original LLM, bypassing gradient obstruction caused by cover layers and ensuring effective gradient propagation during unlearning. Extensive experiments conducted on several unlearning and general capability benchmarks demonstrate that KUnBR achieves state-of-the-art forgetting performance while maintaining model utility.

Beyond Superficial Forgetting: Thorough Unlearning Through Knowledge Density Estimation and Block Re-Insertion

Graph neural networks (GNNs) have revolutionized recommender systems by effectively modeling complex user-item interactions, yet data sparsity and the item cold-start problem significantly impair performance, particularly for new items with limited or no interaction history. While multimodal content offers a promising solution, existing methods result in suboptimal representations for new items due to noise and entanglement in sparse data. To address this, we transform multimodal recommendation into discrete semantic tokenization. We present Sparse-Regularized Multimodal Tokenization for Cold-Start Recommender Systems (MoToRec), a framework centered on a sparsely-regularized Residual Quantized Variational Autoencoder (RQ-VAE) that generates a compositional semantic code of discrete, interpretable tokens, promoting disentangled representations. MoToRec’s architecture is enhanced by three synergistic components: (1) a sparsely regularized RQ-VAE that promotes disentangled representations, (2) a novel adaptive rarity amplification that promotes prioritized learning for cold-start items, and (3) a hierarchical multi-source graph encoder for robust signal fusion with collaborative signals. Extensive experiments on three large-scale datasets demonstrate MoToRec’s superiority over state-of the-art methods in both overall and cold-start scenarios. Our work validates that discrete tokenization provides an effective and scalable alternative for mitigating the long-standing cold-start challenge.

MoToRec: Sparse-Regularized Multimodal Tokenization for Cold-Start Recommender

Recursive transformer (RT) is a promising parameter-sharing technique for reducing computational burden of large-scale model. While RT has been successfully applied to large language models (LLMs), its effectiveness in automatic speech recognition (ASR) remains limited, despite the parallel trend of model scaling in the speech domain. In this paper, we reveal that conventional RT designs for LLMs are suboptimal for speech recognition, primarily because they do not fully consider the layer-wise specialization inherent in the ASR architecture, where lower layers focus on phonetic features and upper layers capture linguistic localization. To address this, we propose BiCycle, a novel RT scheme tailored for ASR. In particular, we firstly analyze attention patterns in a pre-trained ASR model to divide its layers into phonetic and linguistic groups. BiCycle then constructs an efficient RT model by transferring the pre-trained model’s weights in a step-wise manner and applies recursion separately to the phonetic and linguistic groups, preventing conflicts between their roles. Extensive experimental results confirm that the proposed method not only preserves the original ASR mechanism but also outperforms conventional RT approaches.

BiCycle: Group-wise Recursive Transformer Based on ASR Mechanism

Multimodal recommender systems have emerged as a pivotal paradigm for harnessing diverse data modalities to deliver personalized services. Contemporary research predominantly focuses on integrating heterogeneous modality information through graph learning. However, these approaches face two key challenges: (1) the inherent complexity of modalities, characterized by entangled redundant signals and noise; and (2) the challenge of effectively integrating multimodal representations, each of which may exert varying degrees of influence on users' preferences. To address these challenges, we propose a novel Collaboration-Guided $\underline{M}$ultimodal $\underline{D}$isentanglement and $\underline{H}$ierarchical Fusion for $\underline{Rec}$ommendation (DHMRec), which simultaneously achieves intra-modal denoising disentanglement and inter-modal hierarchical fusion. Specifically, we introduce a collaboration-related modality disentanglement module to distinguish between modality-common and modality-specific features. Then, through multi-view graph learning to capture both item-item dependencies and user-item interaction patterns. Additionally, we implement hierarchical fusion between the disentangled multimodal features and ID embeddings using a positive-negative attention-aware fusion module and an interaction distribution-based alignment module. Extensive experiments on three benchmarks demonstrate that our DHMRec surpasses various state-of-the-art baselines, highlighting its effectiveness in intra-modal disentanglement and multimodal features fusion.

Content not yet available

Next from AAAI 2026

Mitigating Noise and Imbalance in Social Governance Graphs for Multi-Type Risk Assessment

Similar lecture

Convolutional LoRA Aggregation for Unseen Tasks Adaptation