OPTD: On-Policy Transition Distillation with Consistency-Guided Adaptive Compression for Few-Step Diffusion Language Models

Diffusion language models (dLLMs) can predict many tokens in parallel, but accurate generation still requires many iterative denoising steps. Few-step distillation accelerates decoding by compressing multiple teacher steps into a single student transition. However, existing methods construct supervision on off-policy trajectories. At inference, the student's early parallel commitments alter the context of later predictions, so the states it actually visits drift away from the supervised ones--precisely when step compression is most aggressive. On-policy distillation is a natural remedy for this mismatch, but it leaves open how far each transition should advance: matching only the teacher's next action limits compression, while indiscriminately merging future actions can violate intermediate dependencies. To address this limitation, we propose OPTD, On-Policy Transition Distillation with consistency-guided adaptive compression. It samples partial states from the few-step student's own trajectories, uses a frozen, question-only teacher to identify outcome-aligned future candidates, and orders them by current-state confidence. The method then selects the longest prefix whose joint commitment preserves the teacher's rollout outcome. A set-bottleneck objective promotes every verified future candidate to the decoder's release threshold, while a frozen-teacher KL anchor regularizes all other active positions. Neither target construction nor training uses a gold response. Across four mathematical reasoning and code-generation benchmarks, OPTD consistently improves the quality--efficiency trade-off and attains the strongest overall quality-constrained AUP among the evaluated few-step baselines.

GrainGS: Gradient-Decoupled Gaussian Splatting for Efficient Dynamic Novel View Synthesis

Dynamic scene reconstruction with 3D Gaussian Splatting requires a balance between fine-grained motion modeling, structural stability, and compact representation. Existing per-primitive methods provide flexible local deformation but often suffer from redundant primitive growth, while anchor-based methods improve spatial regularity at the cost of suppressing locally varying motion. To address these issues, we present GrainGS, a dynamic Gaussian framework that combines a hierarchical anchor scaffold with per-Gaussian deformation. A static warm-up stage first establishes a time-invariant canonical representation from observations across all timestamps. During joint training, a stop-gradient operation blocks the deformation-mediated gradient pathway to the canonical positions while preserving their direct refinement through the reconstruction objective. Each Gaussian then predicts independent temporal offsets for position, rotation, and scale, enabling detailed local motion within a structurally constrained scaffold. A canonical-residual appearance decomposition further models frame-dependent photometric changes without forcing them into geometric deformation. Experiments on synthetic monocular and real-world multiview benchmarks show that GrainGS achieves high reconstruction quality, real-time novel view synthesis, and compact storage. Under the synthetic benchmark setting, it reaches an average peak signal-to-noise ratio of 36.98 decibels, renders at 435.6 frames per second, and requires 4.67 megabytes of storage.

Continual Video-MLLM Adaptation over Evolving Domains

Video multimodal large language models have shown strong capability in video understanding, yet their adaptation to sequentially evolving domains remains underexplored. In real-world deployments, video data often arrives continuously from heterogeneous domains, requiring the model to acquire new domain-specific knowledge without overwriting previously learned capabilities. Existing continual learning methods typically rely on shared adaptation spaces, which can induce severe cross-domain interference and catastrophic forgetting. We propose Distribution-Aware Expert Routing, a parameter-efficient framework for continual Video-MLLM adaptation over evolving domains. DAER maintains domain-isolated lightweight experts while keeping the pretrained Video-MLLM backbone frozen, thereby decoupling domain-specific adaptation from the general multimodal knowledge of the pretrained model. To enable fine-grained specialization, we introduce an intra-domain distribution-aware routing mechanism that matches each input to expert-level prototype reservoirs using MMD. To address the absence of task identities at inference time, we further propose an inter-domain routing mechanism that performs prototype matching in a discriminative subspace for robust domain identification. In addition, we introduce adaptive domain merging to improve parameter scalability and adopt a two-stage optimization strategy to stabilize expert specialization during continual learning. We evaluate DAER by curating a domain-incremental benchmark built from ten VidQA datasets covering diverse visual environments and reasoning demands. Experiments on two strong Video-MLLM backbones show that DAER consistently outperforms prior methods.

Beyond Entropy: Learning from Token-Level Distributional Deviations for LLM Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced Large Language Model (LLM) reasoning; however, it faces a fundamental optimization instability: uniform token updates precipitate entropy collapse, leading to premature convergence to suboptimal strategies, whereas excessive Shannon Entropy maximization can cause entropy explosion, driving blind exploration toward incoherent reasoning chains. To resolve this dichotomy, we introduce the Independent Combinatorial Tokens (ICT) framework, which shifts the optimization focus from scalar uncertainty to the distributional properties of token logits. By leveraging the Jensen-Shannon (JS) divergence between token logits distributions, ICT identifies tokens with distinctive distributional patterns as critical branching points for guiding effective exploration in LLM reasoning. Our theoretical analysis, grounded in both Shannon and second-order Rényi entropy, proves that selectively updating on these tokens regulates policy concentration: it reduces the overall distribution uncertainty measured by Shannon entropy, while controlling probability concentration captured by second-order Rényi entropy. This dual effect prevents over-concentrated token generation from weakening exploration and effectively stabilizes the training landscape. Empirical results demonstrate that updating only the top 10% of unique tokens on Qwen2.5 (0.5B/1.5B/7B) models yields an average pass@4 improvement of 4.58%, with a maximum gain of 14.9%, over GRPO, 20-Entropy, and STAPO baselines across seven benchmarks spanning math, commonsense, and Olympiad-level problems.

Mitigating Bias in Low-SNR Financial Reinforcement Learning via Quantum Representations

The financial market is a typical low signal-to-noise ratio (SNR) setting, which often destabilizes off-policy maximum-entropy methods like Soft Actor-Critic (SAC). Specifically, noisy state representations may produce unreliable Q-value estimates, and bootstrapping amplifies these errors, forming a failure mode we call the "Financial Entropy Trap". In this paper, we propose FPQC-SAC, an efficient and plug-and-play SAC variant that places a compact and bounded Parameterized Quantum Circuit (PQC) before the actor and critic networks to constrain feature propagation at the representation level, rather than filtering raw inputs or regularizing Q-values after bootstrapping. Notably, FPQC-SAC reduces the impact of extreme market fluctuations on Bellman target estimation, while trainable quantum entanglement preserves flexible cross-asset interactions. Empirical evaluations on real-world portfolio management tasks demonstrate that FPQC-SAC substantially enhances out-of-sample stability and cumulative returns by achieving a 66.89% relative gain in cumulative return over standard unconstrained SAC and outperforms the best continuous-control deep reinforcement learning baseline by approximately 27%. Open-source code is available at https://github.com/ZeyuLIU-UST/FPQC-SAC-main.

AdaTok: Self-Budgeting Image Tokenization with Quality-Preserving Dynamic Tokens

Image tokenizers, from 2D grids to recent 1D sequences, typically encode every image with the same fixed number of tokens. Yet visual complexity is highly heterogeneous, so a uniform budget overspends on simple inputs and underserves complex ones. Existing elastic tokenizers expose variable-length reconstructions, but often leave token length as a deployment-time operating point, a search target, or an external prediction rather than an output of the tokenizer itself. In this work, we ask whether a discrete visual tokenizer can budget itself in one pass. Our central finding is that actionable elasticity requires a representation--allocation co-design: prefixes must remain decodable across budgets, and the tokenizer must learn which prefix each image needs. We propose AdaTok, a self-budgeting discrete 1D tokenizer. AdaTok combines Prioritized Representation Learning, which orders tokens with nested tail masking and resolves budget-dependent semantic shift through Multi-Head LoRA decoder heads, with Adaptive Token Allocation, which trains a lightweight deterministic-group GRPO policy over candidate budgets. Dynamic Pareto Weighting balances fidelity and efficiency during policy training without manual trade-off sweeps. On ImageNet-1K, AdaTok-Full reaches rFID 1.31 at 256 tokens, while AdaTok-Adaptive attains rFID 1.50 using only ~118 tokens on average, outperforming discrete 1D baselines at comparable budgets. In autoregressive image generation, the shorter adaptive representation yields ~2.1x throughput over a fixed 256-token decode, suggesting that visual token count can be learned as a content-conditioned output rather than set as a fixed hyperparameter.

Consistent-Inversion: Reverse Consistency Guidance for Structure-Preserving Visual Editing

Text-guided diffusion models have become effective tools for real-image visual editing, where the edited image must follow a target instruction while preserving editing-irrelevant structure. Most training-free editors rely on inversion: a source image is mapped to a noisy latent trajectory and the terminal latent is reused for target-prompt denoising. This reuse is useful for preservation, but it also couples source reconstruction and target editing. The resulting trajectory mismatch may either damage background/layout details or over-constrain the intended edit. This paper presents Consistent-Inversion, a training-free reverse consistency guidance framework for structure-preserving visual editing. Instead of treating the inverted source latent as a fixed initialization, Consistent-Inversion checks whether an intermediate target trajectory can be reversed toward the source inversion trajectory under the source prompt. To make this check well-defined, we construct an auxiliary target-side noise representation, perform source-guided reverse denoising, and use the resulting reverse consistency discrepancy as a correction signal for selected early target denoising steps. The method does not update model parameters, is compatible with inversion-based editors, and introduces only a small inference overhead when applied sparsely. Experiments on PIE-Bench show that Consistent-Inversion improves background and structural fidelity under a unified SD3.5 protocol while maintaining target-prompt alignment, and compatibility experiments further verify the same correction principle on classical Stable-Diffusion inversion pipelines.

STiTch: Semantic Transition and Transportation in Collaboration for Training-Free Zero-Shot Composed Image Retrieval

Training-free zero-shot composed image retrieval models are recently gaining increasing research interest due to their generalizability and flexibility in unseen multimodal retrieval. Recent LLM-based advances focus on generating the expected target caption by exploring the compositional ability behind the LLMs. Although efficient, we find that 1) the generated captions tend to introduce unexpected features from the reference image due to the semantic gap between the input image and text modification, where the image contains much more details than the text; 2) the point-to-point alignment during the retrieval stage fails to capture diverse compositions. To address these challenges, we introduce a novel Semantic Transition and Transportation in collaboration framework for training-free zero-shot CIR tasks. Specifically, given the composed caption inferred by an LLM, we aim to refine it through a transition vector in the embedding space and make it closer to the target image. Combining LLMs with user instruction, the refined caption concentrates more on the core modification intent and thus filters out unnecessary noise. Moreover, to explore diverse alignment during the retrieval stage, we model the caption and image as discrete distributions and reformulate the retrieval task as a set-to-set alignment task. Finally, a bidirectional transportation distance is developed to consider fine-grained alignments across modalities and calculate the retrieval score. Extensive experiments demonstrate that our method can be general, effective, and beneficial for many CIR tasks.

What You Think is What You See: Driving Exploration in VLM Agents via Visual-Linguistic Curiosity

To navigate partially observable visual environments, recent VLM agents increasingly internalize world modeling capabilities into their policies via explicit CoT reasoning, enabling them to mentally simulate futures before acting. However, relying solely on passive reasoning over visited states is insufficient for sparse-reward tasks, as it lacks the epistemic drive to actively uncover the ``known unknown'' required for robust generalization. We ask: Can VLM agents actively find signals that challenge and refine their internal world model through curiosity-driven exploration? In this work, we propose GLANCE, a unified framework that bridges reasoning and exploration by grounding the agent's linguistic world model into the stable visual representations of an evolving target network. Crucially, GLANCE leverages the discrepancy between linguistic prediction and visual reality as an intrinsic curiosity signal within reinforcement learning, steering the agent to actively explore areas where its internal model is uncertain. Extensive experiments across a series of agentic tasks show the effectiveness of GLANCE, and demonstrate that aligning ``what the agent thinks'' with ``what the agent sees'' is key to solving complex or sparse agentic tasks.

GR4CIL: Gap-compensated Routing for CLIP-based Class Incremental Learning

Class-Incremental Learning (CIL) aims to continuously acquire new categories while preserving previously learned knowledge. Recently, Contrastive Language-Image Pre-trained (CLIP) models have shown strong potential for CIL due to their powerful generalization ability. However, existing methods still face two key challenges: shared-parameter adaptation tends to cause old-knowledge drift, and task-specific knowledge organization often leads to poorly calibrated cross-task responses, making reliable routing difficult. To address these issues, we propose GR4CIL, a framework combining task discrimination and knowledge routing for CLIP-based CIL. GR4CIL preserves task-specific visual knowledge while maintaining an incrementally stable shared textual semantic space, thereby reducing interference across tasks. Moreover, we introduce an orthogonal compensation mechanism to mitigate modality-gap-induced bias, enhance within-task discrimination, and enlarge the score margin between the ground-truth task and competing tasks. As a result, GR4CIL enables more reliable task-aware routing over learned knowledge while retaining the zero-shot generalization capability. Experiments on multiple benchmarks show that GR4CIL consistently outperforms strong baselines.