Analytic Subspace Routing: How Recursive Least Squares Works in Continual Learning of Large Language Model

Large Language Models (LLMs) possess encompassing capabilities that can process diverse language-related tasks. However, finetuning on LLMs will diminish this general skills and continual finetuning will further cause severe degradation on accumulated knowledge. Recently, Continual Learning (CL) in Large Language Models (LLMs) arises which aims to continually adapt the LLMs to new tasks while maintaining previously learned knowledge and inheriting general skills. Existing techniques either leverage previous data to replay, leading to extra computational costs, or utilize a single parameter-efficient module to learn the downstream task, constraining new knowledge absorption with interference between different tasks. Toward these issues, this paper proposes Analytic Subspace Routing(ASR) to address these challenges. For each task, we isolate the learning within a subspace of deep layers' features via low-rank adaptation, eliminating knowledge interference between different tasks. Additionally, we propose an analytic routing mechanism to properly utilize knowledge learned in different subspaces. Our approach employs Recursive Least Squares to train a multi-task router model, allowing the router to dynamically adapt to incoming data without requiring access to historical data. Also, the router effectively assigns the current task to an appropriate subspace and has a non-forgetting property of previously learned tasks with a solid theoretical guarantee. Experimental results demonstrate that our method achieves near-perfect retention of prior knowledge while seamlessly integrating new information, effectively overcoming the core limitations of existing methods. Our code will be released after acceptance.

Semantic Shift Estimation via Dual-Projection and Classifier Reconstruction for Exemplar-Free Class-Incremental Learning

Exemplar-Free Class-Incremental Learning (EFCIL) aims to sequentially learn from distinct categories without retaining exemplars but easily suffers from catastrophic forgetting of learned knowledge. While existing EFCIL methods leverage knowledge distillation to alleviate forgetting, they still face two critical challenges: semantic shift and decision bias. Specifically, the embeddings of old tasks shift in the embedding space after learning new tasks, and the classifier becomes biased towards new tasks due to training solely with new data, thereby hindering the balance between old and new knowledge. To address these issues, we propose the Dual-Projection Shift Estimation and Classifier Reconstruction (DPCR) approach for EFCIL. DPCR effectively estimates semantic shift through a dual-projection, which combines a learnable transformation with a row-space projection to capture both task-wise and category-wise shifts. Furthermore, to mitigate decision bias, DPCR employs ridge regression to reformulate classifier training as a reconstruction process. This reconstruction exploits previous information encoded in covariance and prototype of each class after calibration with estimated shift, thereby reducing decision bias. Extensive experiments demonstrate that, across various datasets, DPCR effectively balances old and new tasks, outperforming state-of-the-art EFCIL methods.

RewardDS: Privacy-Preserving Fine-Tuning for Large Language Models via Reward Driven Data Synthesis

The success of large language models (LLMs) has attracted many individuals to fine-tune them for domain-specific tasks by uploading their data. However, in sensitive areas like healthcare and finance, privacy concerns often arise. One promising solution is to sample synthetic data with Differential Privacy (DP) guarantees to replace private data. However, these synthetic data contain significant flawed data, which are considered as noise. Existing solutions typically rely on naive filtering by comparing ROUGE-L scores or embedding similarities, which are ineffective in addressing the noise. To address this issue, we propose RewardDS, a novel privacy-preserving framework that fine-tunes a reward proxy model and uses reward signals to guide the synthetic data generation. Our RewardDS introduces two key modules, Reward Guided Filtering and Self-Optimizing Refinement, to both filter and refine the synthetic data, effectively mitigating the noise. Extensive experiments across medical, financial, and code generation domains demonstrate the effectiveness of our method.

SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddings

Multimodal Large Language Models (MLLMs) have serious security vulnerabilities.While safety alignment using multimodal datasets consisting of text and data of additional modalities can effectively enhance MLLM's security, it is costly to construct these datasets. Existing low-resource security alignment methods, including textual alignment, have been found to struggle with the security risks posed by additional modalities. To address this, we propose Synthetic Embedding augmented safety Alignment (SEA), which optimizes embeddings of additional modality through gradient updates to expand textual datasets. This enables multimodal safety alignment training even when only textual data is available. Extensive experiments on image, video, and audio-based MLLMs demonstrate that SEA can synthesize a high-quality embedding on a single RTX3090 GPU within 24 seconds. SEA significantly improves the security of MLLMs when faced with threats from additional modalities. To assess the security risks introduced by video and audio, we also introduced a new benchmark called VA-SafetyBench. High attack success rates across multiple MLLMs validate its challenge. Our code and data will be available at https://github.com/ZeroNLP/SEA.

PAL: Prompting Analytic Learning with Missing Modality for Multi-Modal Class-Incremental Learning

Multi-modal class-incremental learning (MMCIL) seeks to leverage multi-modal data, such as audio-visual and image-text pairs, thereby enabling models to learn continuously across a sequence of tasks while mitigating forgetting. While existing studies primarily focus on the integration and utilization of multi-modal information for MMCIL, a critical challenge remains: the issue of missing modalities during incremental learning phases. This oversight can exacerbate severe forgetting and significantly impair model performance. To bridge this gap, we propose PAL, a novel exemplar-free framework tailored to MMCIL under missing-modality scenarios. Concretely, we devise modality-specific prompts to compensate for missing information, facilitating the model to maintain a holistic representation of the data. On this foundation, we reformulate the MMCIL problem into a Recursive Least-Squares task, delivering an analytical linear solution. Building upon these, PAL not only alleviates the inherent under-fitting limitation in analytic learning but also preserves the holistic representation of missing-modality data, achieving superior performance with less forgetting across various multi-modal incremental scenarios. Extensive experiments demonstrate that PAL significantly outperforms competitive methods across various datasets, including UPMC-Food101 and N24News, showcasing its robustness towards modality absence and its anti-forgetting ability to maintain high incremental accuracy.

SegACIL: Solving the Stability-Plasticity Dilemma in Class-Incremental Semantic Segmentation

While deep learning has made remarkable progress in recent years, models continue to struggle with catastrophic forgetting when processing continuously incoming data. This issue is particularly critical in continual learning, where the balance between retaining prior knowledge and adapting to new information-known as the stability-plasticity dilemma-remains a significant challenge. In this paper, we propose SegACIL, a novel continual learning method for semantic segmentation based on a linear closed-form solution. Unlike traditional methods that require multiple epochs for training, SegACIL only requires a single epoch, significantly reducing computational costs. Furthermore, we provide a theoretical analysis demonstrating that SegACIL achieves performance on par with joint learning, effectively retaining knowledge from previous data which makes it to keep both stability and plasticity at the same time. Extensive experiments on the Pascal VOC2012 dataset show that SegACIL achieves superior performance in the sequential, disjoint, and overlap settings, offering a robust solution to the challenges of class-incremental semantic segmentation. Code is available at https://github.com/qwrawq/SegACIL.

Grasp What You Want: Embodied Dexterous Grasping System Driven by Your Voice

In recent years, as robotics has advanced, human-robot collaboration has gained increasing importance. However, current robots struggle to fully and accurately interpret human intentions from voice commands alone. Traditional gripper and suction systems often fail to interact naturally with humans, lack advanced manipulation capabilities, and are not adaptable to diverse tasks, especially in unstructured environments. This paper introduces the Embodied Dexterous Grasping System (EDGS), designed to tackle object grasping in cluttered environments for human-robot interaction. We propose a novel approach to semantic-object alignment using a Vision-Language Model (VLM) that fuses voice commands and visual information, significantly enhancing the alignment of multi-dimensional attributes of target objects in complex scenarios. Inspired by human hand-object interactions, we develop a robust, precise, and efficient grasping strategy, incorporating principles like the thumb-object axis, multi-finger wrapping, and fingertip interaction with an object's contact mechanics. We also design experiments to assess Referring Expression Representation Enrichment (RERE) in referring expression segmentation, demonstrating that our system accurately detects and matches referring expressions. Extensive experiments confirm that EDGS can effectively handle complex grasping tasks, achieving stability and high success rates, highlighting its potential for further development in the field of Embodied AI.

Analytic Continual Test-Time Adaptation for Multi-Modality Corruption

Test-Time Adaptation (TTA) aims to help pre-trained model bridge the gap between source and target datasets using only the pre-trained model and unlabelled test data. A key objective of TTA is to address domain shifts in test data caused by corruption, such as weather changes, noise, or sensor malfunctions. Multi-Modal Continual Test-Time Adaptation (MM-CTTA), an extension of TTA with better real-world applications, further allows pre-trained models to handle multi-modal inputs and adapt to continuously-changing target domains. MM-CTTA typically faces challenges including error accumulation, catastrophic forgetting, and reliability bias, with few existing approaches effectively addressing these issues in multi-modal corruption scenarios. In this paper, we propose a novel approach, Multi-modality Dynamic Analytic Adapter (MDAA), for MM-CTTA tasks. We innovatively introduce analytic learning into TTA, using the Analytic Classifiers (ACs) to prevent model forgetting. Additionally, we develop Dynamic Selection Mechanism (DSM) and Soft Pseudo-label Strategy (SPS), which enable MDAA to dynamically filter reliable samples and integrate information from different modalities. Extensive experiments demonstrate that MDAA achieves state-of-the-art performance on MM-CTTA tasks while ensuring reliable model adaptation.

TS-ACL: A Time Series Analytic Continual Learning Framework for Privacy-Preserving and Class-Incremental Pattern Recognition

Class-incremental Learning (CIL) in Time Series Classification (TSC) aims to incrementally train models using the streaming time series data that arrives continuously. The main problem in this scenario is catastrophic forgetting, i.e., training models with new samples inevitably leads to the forgetting of previously learned knowledge. Among existing methods, the replay-based methods achieve satisfactory performance but compromise privacy, while exemplar-free methods protect privacy but suffer from low accuracy. However, more critically, owing to their reliance on gradient-based update techniques, these existing methods fundamentally cannot solve the catastrophic forgetting problem. In TSC scenarios with continuously arriving data and temporally shifting distributions, these methods become even less practical. In this paper, we propose a Time Series Analytic Continual Learning framework, called TS-ACL. Inspired by analytical learning, TS-ACL transforms neural network updates into gradient-free linear regression problems, thereby fundamentally mitigating catastrophic forgetting. Specifically, employing a pre-trained and frozen feature extraction encoder, TS-ACL only needs to update its analytic classifier recursively in a lightweight manner that is highly suitable for real-time applications and large-scale data processing. Additionally, we theoretically demonstrate that the model obtained recursively through the TS-ACL is exactly equivalent to a model trained on the complete dataset in a centralized manner, thereby establishing the property of absolute knowledge memory. Extensive experiments validate the superior performance of our TS-ACL.

C-Adapter: Adapting Deep Classifiers for Efficient Conformal Prediction Sets

Conformal prediction, as an emerging uncertainty quantification technique, typically functions as post-hoc processing for the outputs of trained classifiers. To optimize the classifier for maximum predictive efficiency, Conformal Training rectifies the training objective with a regularization that minimizes the average prediction set size at a specific error rate. However, the regularization term inevitably deteriorates the classification accuracy and leads to suboptimal efficiency of conformal predictors. To address this issue, we introduce \textbf{Conformal Adapter} (C-Adapter), an adapter-based tuning method to enhance the efficiency of conformal predictors without sacrificing accuracy. In particular, we implement the adapter as a class of intra order-preserving functions and tune it with our proposed loss that maximizes the discriminability of non-conformity scores between correctly and randomly matched data-label pairs. Using C-Adapter, the model tends to produce extremely high non-conformity scores for incorrect labels, thereby enhancing the efficiency of prediction sets across different coverage rates. Extensive experiments demonstrate that C-Adapter can effectively adapt various classifiers for efficient prediction sets, as well as enhance the conformal training method.