Singapore

Low-Rank Adaptation (LoRA) is the leading parameter-efficient fine-tuning method for Large Language Models (LLMs), but it still suffers from catastrophic forgetting. Recent work has shown that specialized LoRA initialization can alleviate catastrophic forgetting. There are currently two approaches to LoRA initialization aimed at preventing knowledge forgetting during fine-tuning: (1) making residual weights close to pre-trained weights, and (2) ensuring the space of LoRA initialization is orthogonal to pre-trained knowledge. The former is what current methods strive to achieve, while the importance of the latter is not sufficiently recognized. We find that the space of LoRA initialization is the key to preserving pre-trained knowledge rather than the residual weights. Existing methods like MiLoRA propose making the LoRA initialization space orthogonal to pre-trained weights. However, MiLoRA utilizes the null space of pre-trained weights. Compared to pre-trained weights, the input activations of pre-trained knowledge take into account the parameters of all previous layers as well as the input data, while pre-trained weights only contain information from the current layer. Moreover, we find that the effective ranks of input activations are much smaller than those of pre-trained weights. Thus, the null space of activations is more accurate and contains less pre-trained knowledge information compared to that of weights. Based on these, we introduce LoRA-Null, our proposed method that initializes LoRA in the null space of activations. Experimental results show that LoRA-Null effectively preserves the pre-trained world knowledge of LLMs while achieving good fine-tuning performance, as evidenced by extensive experiments. The code is included in the supplementary material.

AAAI 2026

Put the Space of LoRA Initialization to the Extreme to Preserve Pre-trained Knowledge

knowledge forgetting

parameter-efficient fine-tuning

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

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.

Mixed-precision quantization (MPQ) is crucial for deploying deep neural networks on resource-constrained devices, but finding the optimal bit-width for each layer represents a complex combinatorial optimization problem. Current state-of-the-art methods rely on computationally expensive search algorithms or local sensitivity heuristic proxies like the Hessian, which fail to capture the cascading global effects of quantization error. In this work, we argue that the quantization sensitivity of a layer should not be measured by its local properties, but by its impact on the information flow throughout the entire network. We introduce InfoQ, a novel framework for mixed-precision quantization that is training-free in the bit-width search phase. InfoQ assesses layer importance by performing a single forward pass to measure the change in mutual information in the remaining part of the network, thus creating a global sensitivity score. This approach directly quantifies how quantizing one layer degrades the information characteristics of subsequent layers. The resulting scores are used to formulate bit-width allocation as an integer linear programming problem, which is solved efficiently to minimize total sensitivity under a given budget (e.g., model size or BitOps). Our retraining-free search phase provides a superior search-time/accuracy trade-off (using two orders of magnitude less data compared to state-of-the-art methods such as LIMPQ), while yielding up to a 1\% accuracy improvement for MobileNetV2 and ResNet18 on ImageNet at high compression rates (14.00$\times$ and 10.66$\times$).

InfoQ: Mixed-Precision Quantization via Global Information Flow

Image-guided surgery demands adaptive, real-time decision support, yet static AI models struggle with structured task planning and providing interactive guidance. Large language models (LLMs)-powered agents offer a promising solution by enabling dynamic task planning and predictive decision support. Despite recent advances, the absence of surgical agent datasets and robust parameter-efficient fine-tuning techniques limits the development of LLM agents capable of complex intraoperative reasoning. In this paper, we introduce Surgical AI Copilot, an LLM agent for image-guided pituitary surgery, capable of conversation, planning, and task execution in response to queries involving tasks such as MRI tumor segmentation, endoscope anatomy segmentation, overlaying preoperative imaging with intraoperative views, instrument tracking, and surgical visual question answering (VQA). To enable structured agent planning, we develop the PitAgent dataset, a surgical context-aware planning dataset covering surgical tasks like workflow analysis, instrument localization, anatomical segmentation, and query-based reasoning. Additionally, we propose DEFT-GaLore, a Deterministic Energy-based Fourier Transform (DEFT) gradient projection technique for efficient low-rank adaptation of recent LLMs (e.g., LLaMA 3.2, Qwen 2.5), enabling their use as surgical agent planners. We extensively validate our agent's performance and the proposed adaptation technique against other state-of-the-art low-rank adaptation methods on agent planning and prompt generation tasks, including a zero-shot surgical VQA benchmark, demonstrating the significant potential for truly efficient and scalable surgical LLM agents in real-time operative settings.

Surgical AI Copilot: Energy-Based Fourier Gradient Low-Rank Adaptation for Surgical LLM Agent Reasoning and Planning

Large Vision-Language Models (LVLMs) have recently advanced robotic manipulation by leveraging vision for scene perception and language for instruction following. However, existing methods rely heavily on costly human-annotated training datasets, which limits their generalization and causes them to struggle in out-of-domain (OOD) scenarios, reducing real-world adaptability. To address these challenges, we propose ManipLVM-R1, a novel reinforcement learning framework that replaces traditional supervision with Reinforcement Learning using Verifiable Rewards (RLVR). By directly optimizing for task-aligned outcomes, our method enhances generalization and physical reasoning while removing the dependence on costly annotations. Specifically, we design two rule-based reward functions targeting key robotic manipulation subtasks: an Affordance Perception Reward to enhance localization of interaction regions, and a Trajectory Match Reward to ensure the physical plausibility of action paths. These rewards provide immediate feedback and impose spatial-logical constraints, encouraging the model to go beyond shallow pattern matching and instead learn deeper, more systematic reasoning about physical interactions.

ManipLVM-R1: Reinforcement Learning for Reasoning in Embodied Manipulation with Large Vision-Language Models

Optimising traffic signals is crucial for mitigating urban congestion, and automated planning, particularly with PDDL+, has shown promise for real-world deployment due to its flexibility and centralised perspective. While existing PDDL+ models guarantee deployability on current infrastructure, they face significant limitations: reliance on domain-independent heuristics restricts their applicability and their scalability, and leads to slow solution generation and unclear quality of plans. 

To overcome these challenges and unlock the widespread adoption of planning-based traffic control, we introduce $h^{\text{CAFE}}$, a domain-specific heuristic for PDDL+-based traffic signal optimisation. Unlike prior approaches, $h^{\text{CAFE}}$ is designed to work effectively across multiple problem encodings, addressing a key limitation of traditional domain-specific heuristics. We demonstrate its capabilities on real-world data from a region of the UK, showing significant improvements in solution generation time and search space exploration. Our evaluation also compares the strategies generated by $h^{\text{CAFE}}$ against historical data from existing traffic control systems and a non-deployable benchmark, confirming the high quality of the resulting plans.

A Domain-specific Heuristic for PDDL+-based Traffic Signal Optimisation

Large language models (LLMs) have seen remarkable growth in recent years. To leverage convenient LLM cloud services, users are inevitably to upload their prompts. Additionally, for tasks such as translation, reading comprehension, and summarization, associated files or context are inherently needed, whether or not they contain user privacy information. Despite the rapid progress in LLM capabilities, research on preserving user privacy during inference has been relatively scarce. To this end, this paper conducts some exploratory research in this domain. Firstly, we show that (1) the embedding space of tokens is highly sparse, and (2) LLMs primarily function in the orthogonal subspace of embedding space, these two factors making privacy extremely vulnerable. Then, we analyze the structural characteristics of LLMs and design a distributed privacy-preserving inference paradigm which can effectively resist privacy attacks. Finally, we perform a thorough evaluation of the defended models on mainstream tasks and find that low-bit quantization techniques can be effectively combined with our inference paradigm, achieving a balance between privacy, utility, and runtime memory efficiency.

Reconstruction Attack-Resistant Inference Paradigm for LLM Cloud Services

Adapting large pre-trained language models to downstream tasks often entails fine-tuning millions of parameters or deploying costly dense weight updates, which hinders their use in resource-constrained environments. Low-rank Adaptation (LoRA) reduces trainable parameters by factorizing weight updates, yet the underlying dense weights still impose high storage and computation costs. Magnitude-based pruning can yield sparse models but typically degrades LoRA’s performance when applied naively. In this paper, we introduce SALR (Sparsity-Aware Low-Rank Representation), a novel fine-tuning paradigm that unifies low-rank adaptation with sparse pruning under a rigorous mean-squared-error framework. We prove that statically pruning only the frozen base weights minimizes the pruning error bound, and we recover the discarded residual information via a truncated-SVD low-rank adapter, which provably reduces per-entry MSE by a factor of $(1 - r/\min(d,k))$. To maximize hardware efficiency, we fuse multiple low-rank adapters into a single concatenated GEMM, and we adopt a bitmap-based encoding with a two-stage pipelined decoding+GEMM design to achieve true model compression and speedup. Empirically, SALR attains 50\% sparsity on various LLMs while matching the performance of LoRA on GSM8K and MMLU, reduces model size by $2\times$, and delivers up to a $1.7\times$ inference speedup.

SALR: Sparsity-Aware Low-Rank Representation for Efficient Fine-Tuning of Large Language Models

Confidence alone is often misleading in hyperspectral image classification, as models tend to mistake high predictive scores for correctness while lacking awareness of uncertainty. This leads to confirmation bias, especially under sparse annotations or class imbalance, where models overfit confident errors and fail to generalize. We propose CABIN (Cognitive-Aware Behavior-Informed learNing), a semi-supervised framework that addresses this limitation through a closed-loop learning process of perception, action, and correction. CABIN first develops perceptual awareness by estimating epistemic uncertainty, identifying ambiguous regions where errors are likely to occur. It then acts by adopting an Uncertainty-Guided Dual Sampling Strategy, selecting uncertain samples for exploration while anchoring confident ones as stable pseudo-labels to reduce bias. To correct noisy supervision, CABIN introduces a Fine-Grained Dynamic Assignment Strategy that categorizes pseudo-labeled data into reliable, ambiguous, and noisy subsets, applying tailored losses to enhance generalization. Experimental results show that a wide range of state-of-the-art methods benefit from the integration of CABIN, with improved labeling efficiency and performance.

Perceive, Act and Correct: Confidence Is Not Enough for Hyperspectral Classification

Backdoor attacks embed malicious behaviors into Large Language Models (LLMs), enabling adversaries to trigger harmful outputs or bypass safety controls. However, the persistence of the implanted backdoors under user-driven post-deployment continual fine-tuning has been rarely examined. Most prior works evaluate the effectiveness and generalization of implanted backdoors only at releasing and empirical evidence shows that naively injected backdoor persistence degrades after updates. In this work, we study whether and how implanted backdoors persist through a multi‑stage post-deployment fine‑tuning. We propose P‑Trojan, a trigger‑based attack algorithm that explicitly optimizes for backdoor persistence across repeated updates. By aligning poisoned gradients with those of clean tasks on token embeddings, the implanted backdoor mapping is less likely to be suppressed or forgotten during subsequent updates. Theoretical analysis shows the feasibility of such persistent backdoor attacks after continual fine-tuning. And experiments conducted on the Qwen2.5 and LLaMA3 families of LLMs, as well as diverse task sequences, demonstrate that P‑Trojan achieves over \textbf{99\%} persistence while preserving clean‑task accuracy. Our findings highlight the need for persistence-aware evaluation and stronger defenses in realistic model adaptation pipelines.

Persistent Backdoor Attacks Under Continual Fine-Tuning of LLMs

Zero-shot object navigation tasks agents with locating target objects in unseen environments—a core capability of embodied intelligence. While recent vision-language navigation methods leverage Large Language Models (LLMs) for multimodal reasoning, they suffer from two key limitations: (1) semantic misalignment between language-grounded maps and real-world layouts, and (2) inefficiency due to LLMs’ lack of specialization for navigation-specific tasks. To address these challenges, we propose Chain-of-Search (CoS), a novel parameter-efficient framework that enables human-like decision-making via iterative semantic reasoning. First, CoS replaces traditional global maps with an optimal-benefit multi-map construction that continuously balances expected gain and cost throughout the navigation process. Second, we introduce a Parameter-Efficient Intent Aligner (PEIA), trained via a prompt-guided paradigm to align directional decisions with navigation intent. PEIA injects semantic cues into benefit-aware maps, enabling more rational and goal-consistent exploration. Finally, a Reflection-Guided Destination Verifier (RDV) confirms whether the target is reached via language-driven reasoning and corrects potential errors through self-reflection. CoS achieves state-of-the-art performance on HM3D (+2.8% SR) and MP3D (+1.2% SR) without relying on LLMs, demonstrating the effectiveness of lightweight, reasoning-centered navigation. \textit{All data and code will be publicly released.}

Chain-of-Search: Parameter-Efficient Reasoning for Zero-Shot Object Navigation

In e-commerce logistics, accurate geospatial clustering is essential for optimizing resource allocation, manpower planning, and delivery network design. However, existing density-based clustering approaches, particularly their reliance on heuristic parameter tuning, have been underexplored in datasets with significant density variations, limiting robustness and scalability. This study presents an unsupervised framework that extends DBSCAN by leveraging Gaussian Mixture Models (GMM). First, we propose a method that systematically identifies suitable clustering scales through statistical modeling. Second, the approach iteratively applies DBSCAN to extract clusters from dense to sparse regions, overcoming single-parameter limitations. Finally, we validate the method through large-scale offline experiments using data from over 200 last-mile dispatch centers (LMDC). The results demonstrate the framework’s effectiveness in identifying heterogeneous geographic demand patterns and supporting workforce planning and operational benchmarking. This framework provides a scalable solution to a critical challenge in e-commerce logistics, offering a valuable reference for strategic and operational decision-making.

Content not yet available

Next from AAAI 2026

InfoQ: Mixed-Precision Quantization via Global Information Flow

Similar lecture

LoRA-Flow: Dynamic LoRA Fusion for Large Language Models in Generative Tasks