On the Plasticity and Stability for Post-Training Large Language Models

Training stability remains a critical bottleneck for Group Relative Policy Optimization (GRPO), often manifesting as a trade-off between reasoning plasticity and general capability retention. We identify a root cause as the geometric conflict between plasticity and stability gradients, which leads to destructive interference. Crucially, we argue that deterministic projection methods are suboptimal for GRPO as they overlook the intrinsic stochasticity of group-based gradient estimates. To address this, we propose Probabilistic Conflict Resolution (PCR), a Bayesian framework that models gradients as random variables. PCR dynamically arbitrates conflicts via an uncertainty-aware ``soft projection'' mechanism, optimizing the signal-to-noise ratio. Extensive experiments demonstrate that PCR significantly smooths the training trajectory and achieves superior performance in various reasoning tasks.

Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning

Large language models (LLMs) excel at complex tasks with advances in reasoning capabilities. However, existing reward mechanisms remain tightly coupled to final correctness and pay little attention to the underlying reasoning process: trajectories with sound reasoning but wrong answers receive low credit, while lucky guesses with flawed logic may be highly rewarded, affecting reasoning generalization. From a causal perspective, we interpret multi-candidate reasoning for a fixed question as a family of counterfactual experiments with theoretical supports. Building on this, we propose Group Causal Counterfactual Policy Optimization to explicitly train LLMs to learn generalizable reasoning patterns. It proposes an episodic causal counterfactual reward that jointly captures (i) robustness, encouraging the answer distribution induced by a reasoning step to remain stable under counterfactual perturbations; and (ii) effectiveness, enforcing sufficient variability so that the learned reasoning strategy can transfer across questions. We then construct token-level advantages from this reward and optimize the policy, encouraging LLMs to favor reasoning patterns that are process-valid and counterfactually robust. Extensive experiments on diverse benchmarks demonstrate its advantages.

Bi-C2R: Bidirectional Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification

Lifelong person Re-IDentification (L-ReID) exploits sequentially collected data to continuously train and update a ReID model, focusing on the overall performance of all data. Its main challenge is to avoid the catastrophic forgetting problem of old knowledge while training on new data. Existing L-ReID methods typically re-extract new features for all historical gallery images for inference after each update, known as "re-indexing". However, historical gallery data typically suffers from direct saving due to the data privacy issue and the high re-indexing costs for large-scale gallery images. As a result, it inevitably leads to incompatible retrieval between query features extracted by the updated model and gallery features extracted by those before the update, greatly impairing the re-identification performance. To tackle the above issue, this paper focuses on a new task called Re-index Free Lifelong person Re-IDentification (RFL-ReID), which requires performing lifelong person re-identification without re-indexing historical gallery images. Therefore, RFL-ReID is more challenging than L-ReID, requiring continuous learning and balancing new and old knowledge in diverse streaming data, and making the features output by the new and old models compatible with each other. To this end, we propose a Bidirectional Continuous Compatible Representation (Bi-C2R) framework to continuously update the gallery features extracted by the old model to perform efficient L-ReID in a compatible manner. We verify our proposed Bi-C2R method through theoretical analysis and extensive experiments on multiple benchmarks, which demonstrate that the proposed method can achieve leading performance on both the introduced RFL-ReID task and the traditional L-ReID task.

CKDA: Cross-modality Knowledge Disentanglement and Alignment for Visible-Infrared Lifelong Person Re-identification

Lifelong person Re-IDentification (LReID) aims to match the same person employing continuously collected individual data from different scenarios. To achieve continuous all-day person matching across day and night, Visible-Infrared Lifelong person Re-IDentification (VI-LReID) focuses on sequential training on data from visible and infrared modalities and pursues average performance over all data. To this end, existing methods typically exploit cross-modal knowledge distillation to alleviate the catastrophic forgetting of old knowledge. However, these methods ignore the mutual interference of modality-specific knowledge acquisition and modality-common knowledge anti-forgetting, where conflicting knowledge leads to collaborative forgetting. To address the above problems, this paper proposes a Cross-modality Knowledge Disentanglement and Alignment method, called CKDA, which explicitly separates and preserves modality-specific knowledge and modality-common knowledge in a balanced way. Specifically, a Modality-Common Prompting (MCP) module and a Modality-Specific Prompting (MSP) module are proposed to explicitly disentangle and purify discriminative information that coexists and is specific to different modalities, avoiding the mutual interference between both knowledge. In addition, a Cross-modal Knowledge Alignment (CKA) module is designed to further align the disentangled new knowledge with the old one in two mutually independent inter- and intra-modality feature spaces based on dual-modality prototypes in a balanced manner. Extensive experiments on four benchmark datasets verify the effectiveness and superiority of our CKDA against state-of-the-art methods. The source code of this paper is available at https://github.com/PKU-ICST-MIPL/CKDA-AAAI2026.

Doubly Debiased Test-Time Prompt Tuning for Vision-Language Models

Test-time prompt tuning for vision-language models has demonstrated impressive generalization capabilities under zero-shot settings. However, tuning the learnable prompts solely based on unlabeled test data may induce prompt optimization bias, ultimately leading to suboptimal performance on downstream tasks. In this work, we analyze the underlying causes of prompt optimization bias from both the model and data perspectives. In terms of the model, the entropy minimization objective typically focuses on reducing the entropy of model predictions while overlooking their correctness. This can result in overconfident yet incorrect outputs, thereby compromising the quality of prompt optimization. On the data side, prompts affected by optimization bias can introduce misalignment between visual and textual modalities, which further aggravates the prompt optimization bias. To this end, we propose a Doubly Debiased Test-Time Prompt Tuning method. Specifically, we first introduce a dynamic retrieval-augmented modulation module that retrieves high-confidence knowledge from a dynamic knowledge base using the test image feature as a query, and uses the retrieved knowledge to modulate the predictions. Guided by the refined predictions, we further develop a reliability-aware prompt optimization module that incorporates a confidence-based weighted ensemble and cross-modal consistency distillation to impose regularization constraints during prompt tuning. Extensive experiments across 15 benchmark datasets involving both natural distribution shifts and cross-datasets generalization demonstrate that our method outperforms baselines, validating its effectiveness in mitigating prompt optimization bias.

Hacking Hallucinations of MLLMs with Causal Sufficiency and Necessity

Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities across vision-language tasks. However, they may suffer from hallucinations--generating outputs that are semantically inconsistent with the input image or text. Through causal analyses, we find that: (i) hallucinations with omission may arise from the failure to adequately capture essential causal factors, and (ii) hallucinations with fabrication are likely caused by the model being misled by non-causal cues. To address these challenges, we propose a novel reinforcement learning framework guided by causal completeness, which jointly considers both causal sufficiency and causal necessity of tokens. Specifically, we evaluate each token's standalone contribution and counterfactual indispensability to define a token-level causal completeness reward. This reward is used to construct a causally informed advantage function within the GRPO optimization framework, encouraging the model to focus on tokens that are both causally sufficient and necessary for accurate generation. Experimental results across various benchmark datasets and tasks demonstrate the effectiveness of our approach, which effectively mitigates hallucinations in MLLMs.

Self-Reinforcing Prototype Evolution with Dual-Knowledge Cooperation for Semi-Supervised Lifelong Person Re-Identification

Current lifelong person re-identification (LReID) methods predominantly rely on fully labeled data streams. However, in real-world scenarios where annotation resources are limited, a vast amount of unlabeled data coexists with scarce labeled samples, leading to the Semi-Supervised LReID (Semi-LReID) problem where LReID methods suffer severe performance degradation. Existing LReID methods, even when combined with semi-supervised strategies, suffer from limited long-term adaptation performance due to struggling with the noisy knowledge occurring during unlabeled data utilization. In this paper, we pioneer the investigation of Semi-LReID, introducing a novel Self-Reinforcing Prototype Evolution with Dual-Knowledge Cooperation framework (SPRED). Our key innovation lies in establishing a self-reinforcing cycle between dynamic prototype-guided pseudo-label generation and new-old knowledge collaborative purification to enhance the utilization of unlabeled data. Specifically, learnable identity prototypes are introduced to dynamically capture the identity distributions and generate high-quality pseudo-labels. Then, the dual-knowledge cooperation scheme integrates current model specialization and historical model generalization, refining noisy pseudo-labels. Through this cyclic design, reliable pseudo-labels are progressively mined to improve current-stage learning and ensure positive knowledge propagation over long-term learning. Experiments on the established Semi-LReID benchmarks show that our SPRED achieves state-of-the-art performance. Our source code is available at https://github.com/zhoujiahuan1991/ICCV2025-SPRED

GAPrompt: Geometry-Aware Point Cloud Prompt for 3D Vision Model

Pre-trained 3D vision models have gained significant attention for their promising performance on point cloud data. However, fully fine-tuning these models for downstream tasks is computationally expensive and storage-intensive. Existing parameter-efficient fine-tuning (PEFT) approaches, which focus primarily on input token prompting, struggle to achieve competitive performance due to their limited ability to capture the geometric information inherent in point clouds. To address this challenge, we propose a novel Geometry-Aware Point Cloud Prompt (GAPrompt) that leverages geometric cues to enhance the adaptability of 3D vision models. First, we introduce a Point Prompt that serves as an auxiliary input alongside the original point cloud, explicitly guiding the model to capture fine-grained geometric details. Additionally, we present a Point Shift Prompter designed to extract global shape information from the point cloud, enabling instance-specific geometric adjustments at the input level. Moreover, our proposed Prompt Propagation mechanism incorporates the shape information into the model's feature extraction process, further strengthening its ability to capture essential geometric characteristics. Extensive experiments demonstrate that GAPrompt significantly outperforms state-of-the-art PEFT methods and achieves competitive results compared to full fine-tuning on various benchmarks, while utilizing only 2.19% of trainable parameters. Our code is available at https://github.com/zhoujiahuan1991/ICML2025-VGP.

Componential Prompt-Knowledge Alignment for Domain Incremental Learning

Domain Incremental Learning (DIL) aims to learn from non-stationary data streams across domains while retaining and utilizing past knowledge. Although prompt-based methods effectively store multi-domain knowledge in prompt parameters and obtain advanced performance through cross-domain prompt fusion, we reveal an intrinsic limitation: component-wise misalignment between domain-specific prompts leads to conflicting knowledge integration and degraded predictions. This arises from the random positioning of knowledge components within prompts, where irrelevant component fusion introduces interference.To address this, we propose Componential Prompt-Knowledge Alignment (KA-Prompt), a novel prompt-based DIL method that introduces component-aware prompt-knowledge alignment during training, significantly improving both the learning and inference capacity of the model. KA-Prompt operates in two phases: (1) Initial Componential Structure Configuring, where a set of old prompts containing knowledge relevant to the new domain are mined via greedy search, which is then exploited to initialize new prompts to achieve reusable knowledge transfer and establish intrinsic alignment between new and old prompts. (2) Online Alignment Preservation, which dynamically identifies the target old prompts and applies adaptive componential consistency constraints as new prompts evolve. Extensive experiments on DIL benchmarks demonstrate the effectiveness of our KA-Prompt. Our source code is available at https://github.com/zhoujiahuan1991/ICML2025-KA-Prompt

Vision Graph Prompting via Semantic Low-Rank Decomposition

Vision GNN (ViG) demonstrates superior performance by representing images as graph structures, providing a more natural way to capture irregular semantic patterns beyond traditional grid or sequence-based representations. To efficiently adapt ViG to downstream tasks, parameter-efficient fine-tuning techniques like visual prompting become increasingly essential. However, existing prompting methods are primarily designed for Transformer-based models, neglecting the rich topological relationships among nodes and edges in graph-based representations, limiting their capacity to model complex semantics. In this paper, we propose Vision Graph Prompting (VGP), a novel framework tailored for vision graph structures. Our core insight reveals that semantically connected components in the graph exhibit low-rank properties. Building on this observation, we introduce a semantic low-rank prompting method that decomposes low-rank semantic features and integrates them with prompts on vision graph topologies, capturing both global structural patterns and fine-grained semantic dependencies. Extensive experiments demonstrate our method significantly improves ViG's transfer performance on diverse downstream tasks, achieving results comparable to full fine-tuning while maintaining parameter efficiency. Our code is available at https://github.com/zhoujiahuan1991/ICML2025-VGP.