Distillation 进展

On-policy distillation (OPD) applies token-level teacher supervision to student-generated trajectories, but this supervision is not always reliable. Existing methods use local confidence or teacher-student agreement to weight, filter, or truncate the sampled trajectory. These signals do not directly determine whether the teacher can continue a student prefix to a correct answer, and trajectory-level interventions can conflate one rollout's unreliability with low expected training value of its prompt. We define prompt-level teacher continuation reliability $R$ as the teacher's probability of reaching a correct answer from a student prefix, averaged over prefixes and trajectories induced by the current student. Oracle experiments show that high-$R$ prompts yield larger OPD gains and that descending-$R$ training outperforms random and ascending orders on a fixed prompt pool. Because estimating $R$ requires many teacher continuations, we use the maximum ROUGE-5 F1 between one independent student rollout and verifier-correct same-prompt teacher trajectories. Across ten equal-frequency bins of this actual score, mean $R$ rises monotonically, showing that the proxy separates coarse reliability levels. ReOrder-OPD sorts prompts by the proxy, then draws independent on-policy training trajectories for vanilla OPD. It improves every matched aggregate comparison across Qwen3 and Gemma4 mathematics settings and Qwen3 code settings. Gains in all six FiRe-OPD and ExOPD settings show that prompt ordering complements within-trajectory supervision.

Streaming video generation holds strong potential for world modeling, where future frames must be inferred online sequentially to form a continuous video stream. However, streaming video diffusion models introduce a fundamental train-inference mismatch: inference follows a specialized denoising order, whereas advanced training strategies typically require diverse noise-level configurations. To address this trade-off between train-inference consistency and training coverage, we reformulate the video diffusion sampling as a frame-indexed stochastic process over noise levels. Within this stochastic process space, we construct a continuous training trajectory along which the sampling schedule progressively evolves from independent sampling to inference-consistent sampling. We further introduce a joint calibration algorithm and a temporal correlative sampling algorithm to ensure trajectory smoothness and cross-frame correlation. Building on these designs, we propose Stream Forcing, a unified training framework for streaming video generation that balances training sufficiency and inference efficiency. Extensive experiments demonstrate that Stream Forcing significantly improves generation quality with a 36.6% FVD improvement on the UCF-101 benchmark. Furthermore, our method facilitates robust zero-shot extrapolation to long-horizon video generation with a 27.9% FVD improvement on the UCF-101 benchmark.

Text-to-Video (T2V) generative models are vulnerable to jailbreak attacks in real-world deployment, leading them to produce harmful or inappropriate content. Existing defense approaches mainly rely on input filtering or reconstruction, which not only incur high computational latency but also tend to distort semantics. To address these issues, we experimentally and systematically analyze the differences between clean and jailbreak samples in the cross-attention feature space, revealing for the first time a cumulative separation effect and a progressively increasing trend of linear separability between the two during the diffusion process. Based on this insight, we propose SafeCA, a feature-level defense mechanism for safe cross-attention localization and regularization. Firstly, we identify key defensive regions and values through attention stability analysis using cross-attention features collected from clean prompts within a single inference. Secondly, SafeCA mitigates anomalous activations via attention masking with energy normalization and introduces a lightweight semantic-space adapter to redirect abnormal semantic flows. Furthermore, we detect and suppress potentially malicious tokens by back-propagating feature anomaly signals to the input cue words, thereby enhancing the deployability of the defense in commercial models. Experimental results show that SafeCA reduces the jailbreak success rate by about 20% on mainstream T2V models, adds almost no inference overhead (+0.1s), and maintains good text-video semantic consistency. Overall, SafeCA provides an architecture-level, deployable protection paradigm for T2V generation models.

Omni-modal dialogue models can understand multimodal inputs and synthesize spoken replies, yet their responses remain visually disembodied. We introduce Ex-Omni-2D, an omni-modal dialogue framework that generates a coordinated response comprising text, personalized speech, and reference-conditioned video. Given a multimodal query, reference image, and reference audio, the model predicts a structured Visual Thought Plan (VTP) describing scene, emotion, and motion, followed by response text and native multi-codebook speech units. These units form a shared acoustic-temporal interface: they are decoded into speech and aligned online with video frames. This interface enables the response and avatar pathways to be learned from heterogeneous speech, dialogue, and avatar-video data, avoiding the need for large-scale query--text--speech--video supervision. A full-sequence Video Generator serves as the primary Teacher. For efficient incremental generation, we further distill it into a few-step block-causal Streaming Student whose Prefix Streaming mechanism carries a clean latent across consecutive chunks to reduce cumulative late-chunk degradation. With four-step inference, the complete four-GPU pipeline achieves an end-to-end RTF of 1.293 at $400\times720$/$720\times400$, providing a practical quality--efficiency operating point.

Quantization-Aware Training (QAT) enables the deployment of quantized models with minimal accuracy degradation. However, in practical scenarios, training labels are often unavailable due to privacy, copyright, or cost constraints. Knowledge Distillation (KD) is a common approach to address this challenge, but we observe that prior work combining QAT with KD suffers from a fundamental limitation: during distillation, the range mismatch between the teacher and the quantized student model induces an unattainable residual, resulting in an irreducible lower bound on the distillation loss. Motivated by this observation, we propose SQuaT (Student-Aware Quantized Teacher Features), a label-free QAT framework with KD that theoretically eliminates this lower bound by applying the student's quantization parameters to quantize the teacher's features during distillation. Through comprehensive experiments across diverse settings, we demonstrate that SQuaT consistently outperforms strong baselines, with particularly pronounced gains in extreme low-bit (e.g., 1- and 2-bit) settings. Furthermore, extensive evaluations across various model design choices show that our approach does not rely on specific architectural assumptions, making it broadly applicable across diverse architectures and quantization settings. The source code is available at https://github.com/lcdbsa522/SQuaT.

Audio-visual interaction is the standard for patient-physician consultations, enabling natural communication and effective assessment of illness through non-verbal cues. While text-based AI has shown promise, it discards essential perceptual dimensions and limits patients who cannot articulate symptoms in writing. Early efforts to extend medical AI to audio-visual interaction have demonstrated feasibility but not reached clinician-level performance. Here, we provide the first demonstration of expert-level AI in real-time clinical video consultations using AMIE (Articulate Medical Intelligence Explorer) in a video configuration. AMIE (Video) is a Gemini-based multi-agent system integrating low-latency dialogue, clinical reasoning, and real-time audio-visual perception. To guide development, we established a taxonomy and automated evaluations for clinical audio-visual cues in telehealth settings. In a randomized Objective Structured Clinical Examination (OSCE) study with 30 primary care physicians (PCPs), 15 patient actors and 100 clinical scenarios, we compared AMIE (Video), its text-only counterpart AMIE (Text), and PCPs consulting via video. Clinical evaluators rated AMIE (Video) on par or better than PCPs in history-taking, diagnosis, management, and physical observation and examination. Patient actors preferred AMIE's approach to assessing and explaining conditions, while PCPs were preferred for rapport and partnership building. In modality ablation, patient actors preferred AMIE (Video)'s interface over text chat for communicative effectiveness, convenience, and feeling understood. Limitations remain in fine anatomical precision, subtle affective nuances, and high-frequency movements. While further research is needed before real-world translation, these results mark an important milestone toward AI systems capable of augmenting care across the sensory complexity of clinical practice.

Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63.0-68.0% of groups in our experiments. We propose SKALD (Skill-Anchored Latent Distillation), an on-policy self-distillation framework that uses two context views of the same Qwen3-Base model: a question-only student and a teacher conditioned on an abstract, explicit-answer-filtered skill card. The student is trained on its own prefixes, transferring the skill-induced advantage into shared parameters without privileged input at test time. To stabilize context-induced distribution mismatch, SKALD employs an annealed exponentially tilted objective that downweights teacher-preferred tokens with very low student likelihood; as the tilt vanishes, it converges to teacher cross-entropy and recovers the forward-KL student gradient. An empirical gate activates distillation only when verified rollouts estimate a positive teacher advantage. Across five held-out mathematics benchmarks, SKALD improves overall avg@8 over GRPO by +2.46, +4.85, and +12.01 at 0.6B, 1.7B, and 4B, respectively. At 1.7B, zero-variance-only distillation recovers 84.7% of the full gain, while SKALD remains +4.06 above FLOP-matched GRPO and exceeds contextual skill exposure by +3.77. These results show that abstract skills provide dense supervision where group-relative rewards become uninformative.

On-policy distillation (OPD) has emerged as a core component of modern LLM post-training pipelines, yet we reveal a failure mode: degenerate agreement, where students exploit repetitive loops to achieve near-perfect token agreement with the teacher despite globally flawed responses. We therefore shift our focus from agreement to teacher-student mismatch, and find that mismatch tokens can be mainly categorized into two types: student-excess tokens and student-deficit tokens. Student-excess tokens are generated by the student but assigned near-zero probability by the teacher; their log-ratio corrections grow unbounded and destabilize the update. Student-deficit tokens, in contrast, are preferred by the teacher but rarely sampled by the student; their absence blocks the transfer of the teacher's reasoning patterns. To tackle these mismatch directions, we propose TIDE (Token-level Independent Deficit-Excess correction), which applies bounded Hellinger shaping to suppress the most severe sampled excesses and an analytic teacher top-$K$ injection to restore deficient probability mass without requiring deficit tokens to be sampled. Across mathematical reasoning benchmarks with multiple Qwen3 teacher-student pairs, TIDE consistently outperforms standard OPD and recent token-selection and reward-shaping baselines. Moreover, the gains of TIDE are more pronounced under strong teacher-student mismatch, where it improves Avg@8 from 6.9% to 20.3%, reduces average response length by a factor of 3.6, and substantially reduces formatting failures. Code is available at https://github.com/yzc-666/TIDE

Radio maps describe how wireless signals propagate across space and are essential for wireless communication, sensing, and network planning. However, constructing accurate radio maps traditionally requires either dense measurements or computationally expensive physical simulations, which limits scalability and real-time deployment. Recent advances in generative artificial intelligence offer a promising alternative, but existing approaches lack fine-grained control and physical consistency when applied to real-world wireless environments. Here we present ControlRadio, a controllable generative framework that produces radio maps from natural-language descriptions and environmental layouts, including building structures and transmitter locations. Joint semantic and spatial conditioning enables interpretable, propagation-plausible generation, while a controlled latent prior and layout-aware conditioning improve stability and structural consistency. Extensive experiments demonstrate that ControlRadio achieves state-of-the-art accuracy and strong generalization across diverse urban scenarios, while reducing computation time by more than four orders of magnitude compared with conventional simulation-based methods. Such results suggest a new paradigm for scalable and controllable wireless environment modeling, with broad implications for next-generation communication systems and data-driven radio sensing.

Real-world image super-resolution (Real-ISR) aims to reconstruct high-quality (HQ) images from low-quality (LQ) inputs subject to diverse real-world degradations. Recent advances have leveraged the LQ inputs and natural image priors learned by Stable Diffusion models to achieve impressive results. However, existing methods often overlook insufficient clarity of LQ inputs inevitably induce content drift in the generated HQ images. This manifests primarily as visual detail degradation and textual semantic shift, severely compromising both fidelity and perceptual quality. To address this challenge, we propose FSP-Diff, a novel one-step diffusion model featuring a dual-pathway architecture. This architecture comprises a Detail-Conditioned Pathway for injecting structured details to recover fine structures, and a Detail-Modulated Semantic Pathway that refines semantic guidance using structured details to mitigate semantic deviations. Extensive experiments on standard Real-ISR benchmarks demonstrate that FSP-Diff surpasses existing one-step diffusion methods in both quantitative and qualitative metrics.

On-Policy Self-Distillation (OPSD) is commonly interpreted as the transfer of privileged information: a teacher observes the verified solution to the target problem and supervises the student's trajectory. However, this interpretation conflates two effects. The reference solution not only reveals the answer to the current instance but also changes the context under which the teacher provides token-level supervision. We investigate the role of target-specific privilege with $\mathrm{OP}^{2}\mathrm{SD}$ (On-Policy Self-Distillation from Other Problems), which replaces the paired reference with a problem and solution from a different example, while preserving the student rollout, teacher, and distillation objective. Across three models and three mathematics benchmarks, $\mathrm{OP}^{2}\mathrm{SD}$ improves over the base model, remains competitive with OPSD. The success of $\mathrm{OP}^{2}\mathrm{SD}$ implies that OPSD gains do not necessarily come from access to the reference solution, and that the teacher's context-induced behavior is an important factor.

On-policy distillation (OPD) aligns a student with a teacher on trajectories sampled from the student itself, reducing the train-test state mismatch of offline distillation. The same feedback loop can nevertheless be unstable: each update changes both the policy and the states on which the next update is computed. We introduce WDL-OPD, a mixture-constrained co-training method with two trainable policies. An anchor policy generates every rollout, an auxiliary policy evaluates the same visited states, and a geometric mixture of their token distributions is matched to a frozen teacher by reverse KL. Both policies receive gradient. We show that freezing the auxiliary recovers an anchor-plus-contrast proxy target closely related to OPD$^2$ and W2S-OPD, whereas joint training creates branch-level degrees of freedom that a static delta cannot express. In recorded Qwen3 experiments at 1.7B and 4B scale, WDL-OPD produces the strongest student checkpoint in each of four scale-domain settings. It raises MATH500 accuracy from 0.630 to 0.685 at 4B and from 0.521 to 0.585 at 1.7B. In code generation, seven single-policy OPD configurations exhibit entropy growth or trajectory degradation, while co-training reaches independently re-evaluated development scores of 0.637 and 0.375. Because several comparisons differ in curriculum or initialization, these results support a stabilization hypothesis rather than a universal causal claim. We provide the exact training algorithm, failure evidence, and the controlled comparison matrix needed to test that hypothesis.

On-policy self-distillation (OPSD) converts feedback into dense token-level supervision on trajectories generated by the policy to be optimized, providing a useful complement to reinforcement learning with sparse outcome rewards. However, the self-teacher policy used in OPSD is typically a stop-gradient or exponential-moving-average copy of the policy conditioned on additional context information, and thus co-evolves with both the student policy and its on-policy context distribution. Directly matching such a moving target with a fixed projection objective can lead to unstable optimization or excessive distributional concentration. This nature of OPSD motivates the proposed Self-Referenced On-Policy Self-Distillation (SR-OPSD). At fixed student-generated contexts, a token-level variational characterization identifies the effective distillation target as a geometric interpolation between the self-teacher policy and a reference policy. Meanwhile, we use the Rényi divergence family to generalize the projection geometry. This formulation separates where the adaptive target is placed from how the student is projected toward it: the interpolation coefficient controls underlying target, while the Rényi order controls the projection geometry and its sensitivity to token-level density ratios. Extensive experiments across scientific evaluation, mathematical reasoning, and coding generation tasks with multiple large language models show that SR-OPSD achieves the state-of-the-art or competitive performance across various settings.

Diffusion-based super-resolution (SR) achieves strong perceptual quality but requires costly iterative denoising. Existing one-step distillation methods reduce inference time but depend on expensive pretrained teachers, whereas CTMSR avoids distillation through PF-ODE consistency training yet does not explicitly model the restoration dynamics from low-resolution (LR) inputs to high-resolution (HR) images. We propose MeanSR, a one-step perceptual SR method that learns an LR-conditioned average velocity field to directly capture the finite-time transition from degraded or noisy inputs to plausible HR outputs. We further reformulate distribution trajectory matching for average-velocity generation and introduce a Stage-Aware Temporal Sampling strategy to improve trajectory learning. Experiments on synthetic and real-world benchmarks show that MeanSR outperforms CTMSR on CLIPIQA, MUSIQ, and MANIQA while substantially reducing FLOPs and inference latency. MeanSR also reconstructs sharper structures and more realistic textures with fewer perceptual artifacts.

Flow-matching models are now a mainstream method to image generation, but its adaptation to diverse downstream scenarios typically relies on post-training, which may cause conflicts among task-specific optimization objectives. Reinforcement learning enables direct optimization of task-specific rewards beyond the original models, yet trajectory-level optimization may incur high-variance gradients and cross-task interference. On-policy distillation (OPD) offers dense and stable supervision on student rollouts, but conventional teacher matching remains imitation-based. We propose DreOPD, a Degraded-reference extrapolative OPD method for flow-matching models that bridges these two paradigms. Our DreOPD converts implicit reward extrapolation into closed-form velocity regression, enabling extrapolative post-training with the stability of OPD. It further uses a mildly degraded reference to strengthen the teacher-reference contrast, yielding a clearer extrapolation direction. Experiments on single- and multi-teacher settings show that DreOPD outperforms OPD and multi-task RL baselines in average performance, while surpassing specialized teachers on most metrics.

Image super-resolution (SR) with large generative models has recently achieved remarkable perceptual quality, yet maintaining fidelity to the LR observation remains challenging. In particular, we observe that diffusion transformers (DiTs) built on latent representations suffer from a critical limitation: the compression bottleneck of the VAE weakens fine-grained spatial information, leading to hallucinated details that are weakly grounded in the input image. In this work, we revisit generative SR from a representation perspective and propose a pixel-grounded super-resolution (PGSR) framework that preserves LR-observed pixel evidence before VAE compression and reuses it throughout restoration. Instead of relying solely on the compressed latent condition, PGSR extracts pre-VAE pixel evidence from the upsampled LR image and reuses it at two stages. First, Condition-Side Trajectory Guidance fuses LR-derived pixel evidence with the latent LR condition to guide the latent restoration trajectory. Second, Decoder-Side Pixel Grounding injects multi-scale pixel features into the frozen VAE decoder to ground the final rendering with LR-observed cues. To efficiently adapt large pretrained DiT models, we keep the latent autoencoder and main flow-matching backbone frozen, and train only lightweight restoration modules. We further study an efficient local-window attention variant for improved high-resolution efficiency and scalability. Extensive experiments demonstrate that PGSR improves the realism--fidelity trade-off and produces more faithful, visually convincing results than existing latent generative SR approaches.

Diffusion models have enabled high-quality video generation in recent years, but the high cost of iterative sampling hinders their practical deployment. Few-step distillation alleviates this cost, yet exposes a quality--diversity trade-off between its two dominant paradigms: trajectory-level distillation (e.g., sCM) favors diversity, whereas distribution-level distillation (e.g., DMD) favors quality. Targeting extreme two-step video generation, we introduce DUET, which reconciles the two paradigms through a noise-level duet of experts: an sCM expert takes the high-noise step to lay out diverse structure, and a DMD expert takes the low-noise step to refine appearance detail. Since the two experts are trained independently with their native objectives, DUET sidesteps the optimization difficulties of loss-level combinations and delivers quality and diversity jointly rather than trading one for the other. We further identify the relay interface and the high-noise stage as the remaining bottlenecks, and address them with RL-guided expert adaptation, yielding DUET+. With the Wan2.1-T2V-1.3B backbone, DUET lifts the two-step quality of sCM close to the level of DMD while retaining nearly all of its structural diversity---about twice that of DMD---and DUET+ further improves overall quality while preserving this diversity advantage. Together, these results establish noise-level expert specialization as a simple, effective paradigm for reconciling diversity and quality in two-step video generation.

On-policy self-distillation (OPSD) uses a privileged teacher to supervise a reasoning model on prefixes sampled from its own rollouts. Yet each rollout also reveals how the student's response unfolds and whether it succeeds, student-specific hindsight that standard OPSD does not use to form the teacher. We introduce Privileged Adaptation from Student Trajectories (PAST), which treats each completed student trajectory as additional privileged information for the OPSD teacher while leaving the student's distillation prefixes unchanged. PAST preserves the student's next-token distribution on correct trajectories and uses failed trajectories to adapt the teacher toward verified success under student-proximity regularization. We characterize what such a trajectory-conditioned teacher can transfer to a prefix-only student. Forward-KL distillation projects the teacher distributions to their conditional arithmetic mean given the prefix. This projection separates trajectory-specific variation that remains privileged from the mean policy shift available to the student. For correct trajectories, the unclipped population objective also has the frozen student as an ideal distributional fixed point. Across three mathematical reasoning benchmarks, PAST improves the Avg@12 macro average over Vanilla OPSD by 5.6 percentage points. A $2\times2$ factorial study shows gains from both complete-trajectory access and teacher adaptation, while trajectory removal and shuffling confirm that the adapted teacher uses the matching hindsight context.

On-policy self-distillation improves language-model reasoning by querying a teacher on states actually visited by the student. Recent methods create a powerful information asymmetry by exposing the teacher to privileged context, yet they fundamentally rely on external supervision---such as gold solutions or verifiers---to construct this advantage. We introduce CoDA (Consensus and Disagreement Alignment), a fully unsupervised framework that creates reliable privileged information entirely from the latent uncertainty structure of a model's own unlabeled rollouts. CoDA extracts two complementary signals. In the positive branch, answer-level consensus identifies a stable reasoning mode, which conditions a frozen self-teacher to provide dense distributional guidance on fresh student trajectories. However, because agreement does not guarantee correctness, positive-only distillation risks amplifying correlated errors into a false consensus. To break this harmful feedback loop, CoDA incorporates a negative branch that exploits disagreement: minority trajectories are treated as unstable alternatives and gently penalized via a reference-anchored, KTO-style calibration objective. This unpaired binary feedback provides robust regularization without requiring the strong assumption that the consensus is the absolute ground truth. Empirical evaluations on competition-level mathematical benchmarks demonstrate that CoDA significantly improves reasoning, outperforming self-generated baselines and effectively stabilizing training against erroneous consensus.

Zero-shot text-to-speech (TTS) now supports interactive assistants, personalized media, and accessibility tools. All TTS systems require faithful linguistic rendering, consistent speaker identity, and low-latency response. Yet compact streaming systems must preserve sufficient acoustic detail in a predictable low-rate latent sequence, while iterative diffusion sampling and classifier-free guidance multiply inference cost at every autoregressive step. To strike a balance between high-fidelity synthesis and low-latency inference, we present CuteTTS, a compact continuous-autoregressive TTS system. It combines semantically aligned causal VAE latents with patch-level autoregression, explicit speaker conditioning, and a bidirectional flow-matching head. We further introduce guidance-step distillation, which absorbs classifier-free guidance and multiple solver steps into a single interval-conditioned student. Evaluations on LibriSpeech and Seed-TTS-Eval demonstrate competitive intelligibility and speaker similarity in zero-shot voice cloning, while distillation lowers first-audio latency by 23.3% and real-time factor by 40.8% relative to the base model with comparable objective and subjective quality. These results provide a practical path toward continuous-autoregressive TTS that reconciles high-fidelity generation with the latency demands of real-time interaction.

Traffic data collection is dominated today by deep object detectors followed by tracking-by-detection, a pipeline that presupposes what is often missing in practice: a detector already trained on the class one wants to count. We revisit a purely geometric traffic-sensing pipeline for Single Board Computers in which detection is class-agnostic: moving objects come from background subtraction and thresholding, and counting is decided by a geometric rule on an imaginary line across the road, a software inductive loop detector. With no object model, training set or per-object trajectory, it runs faster than real time on Raspberry Pi class hardware. Two counting rules are described: a constant average speed rule, whose expected accuracy is derived analytically as about 86% under a Gaussian speed distribution, and a self-calibrating pre-calibration rule that recovers the lane geometry from blob statistics and counts edges of lane occupancy, additionally yielding per-vehicle average speed at no extra cost. Over four videos the latter counts with 83.3%-100% accuracy; in a field deployment it reaches 91% against 37.5% for a blob-tracking baseline under the same compute budget. We report the observations of that period in detail: the resolution floor below which accuracy collapses, the frame rate floor at which vehicles alias past the counting line, the gap between short curated clips and long uncontrolled footage, and the trade-off between Python (easier to tune, 100% CPU) and C++ (40% CPU, thermally viable). These are properties of the sampling geometry, not of the hardware of the time, and still constrain edge deployments. We close by arguing where motion-based, class-agnostic detection remains the right tool: open-set classes with no annotated data, tight power budgets, privacy-constrained installations, and the cold start of mining training crops to bootstrap a learned detector.

We present InsertFuse, a unified framework for multi-category reference-guided image insertion. Its key idea is to decouple category-specific expertise learning from cross-category capability consolidation. InsertFuse first trains specialized experts for different insertion categories and then introduces Insertion On-Policy Distillation (IOPD) to consolidate their capabilities into a single student. By querying the matched expert at states visited by the student, IOPD preserves category-specific insertion behavior while mitigating the cross-category interference caused by direct joint training. To improve spatial control, we propose Token-Aligned Geometry Conditioning (TAGC), which maps mask-derived geometric cues to the visual token grid, and Region-Balanced Flow Matching, which separately normalizes prediction errors inside and outside the insertion region to prevent background-dominated and scale-dependent supervision. We further introduce Reference CFG to isolate and strengthen the guidance induced by the visual reference under fixed scene and geometry conditions, with IOPD transferring this enhanced supervision into the unified student. Extensive experiments on the public AnyInsertion benchmark and our multi-category test set demonstrate state-of-the-art performance on most metrics, showing strong reference fidelity and generation quality across diverse insertion categories.

On-policy (Self-)Distillation (OPD / OPSD) has shown strong potential for post-training large language models (LLMs). However, existing methods still rely heavily on external supervision, including ground-truth signals, environmental feedback, or guidance from larger models, and therefore fall short of genuine "self"-distillation. In this study, we show that on-policy self-distillation can be achieved using only a model's own generations via internal consistency. We propose unsupervised on-policy self-distillation (U-OPSD). U-OPSD first samples multiple rollouts and constructs a pseudo solution by majority vote under a self-consistency threshold. It then conditions the model's distribution on the pseudo-solution and distills itself on the disagreeing completions, allowing the model to correct itself precisely where it is confidently wrong. Across diverse benchmarks, base models, and training settings, U-OPSD consistently improves over the base models and matches or surpasses supervised methods with ground truth (GT) such as OPSD and GRPO. On five mathematical reasoning benchmarks, i.e., AIME24, AIME25, HMMT25, MATH500, and AMC23, U-OPSD improves over the base model by 8.5% and 10.7% on Qwen3 non-thinking mode at 4B and 8B scales, and outperforms OPSD by 3.2% and 2.3% on average, respectively. In thinking mode, U-OPSD stays on par with OPSD, ahead by 0.9% at 4B and level at 8B and surpassing GRPO by 0.7% and 1.1%, respectively. Code is available at [https://github.com/williamium3000/u-opsd](https://github.com/williamium3000/u-opsd).

Multilingual reasoning transfer is crucial for extending reasoning capabilities of large language models (LLMs) beyond high-resource languages. On-policy self-distillation (OPSD) and its variants have emerged as a promising paradigm, providing dense token-level supervision on student-generated rollouts, yet their objectives do not explicitly prioritize reasoning signals most critical to cross-lingual transfer. We characterize that target-language reasoning comprises the generation of both surface text and reasoning pivots, which are decisions that advance or redirect the reasoning process and shape subsequent inference. This motivates concentrating privileged distillation around such pivots. We therefore propose RP-OPSD, Reasoning-Pivot-guided On-Policy Self-Distillation, using the distributional shift between matched teacher views with and without an English reference solution as an operational proxy to guide privileged distillation and reference anchoring. Experiments on mathematical reasoning benchmarks covering 17 languages and multiple difficulty levels show that our method outperforms strong multilingual reasoning baselines and OPSD variants. Further analysis reveals that RP-OPSD concentrates privileged distillation on reasoning-control and problem-condistioned state-update tokens, while downweighting it for tokens that mainly support surface realization. Our code is available at https://github.com/NJUNLP/RP-OPSD.

On-Policy Self-Distillation (OPSD) has become a standard post-training approach for improving visual reasoning in multimodal large language models (MLLMs). Existing methods draw privileged information from diverse input sources to guide self-distillation. Yet these designs overlook Modality Imbalance, a challenge inherent to MLLM reasoning. When textual information dominates generation, the model cannot fully integrate its multimodal input. Consequently, carefully designed privileged information remains underused, limiting the effectiveness of OPSD. To examine this limitation, we construct a Positive Teacher with the Zoom-In Image and a Negative Teacher with the Mask Image, which exhibit different degrees of Modality Imbalance. Changes in their reasoning correctness and token logits reveal that Modality Balance can itself serve as privileged information. Motivated by this finding, we introduce OPD-V, a visual OPSD paradigm that instantiates such information through the Positive Teacher and Negative Teacher. Positive Modality-Balance Logits Margins define a Modality-Balance Trust Region that selects the on-policy tokens used for self-distillation. Experiments across 6 benchmarks, 4 MLLM backbones, and 5 post-training methods show that OPD-V consistently improves reasoning performance while reducing training cost.

Vision-based guidance of unmanned aerial vehicles (UAVs) toward unmanned ground vehicles (UGVs) supports cooperative aerial--ground robotics, but reliable continuous yaw estimation from onboard vision remains challenging because of sensing uncertainty, limited computation, and the need for interpretable control. Existing deep-learning and geometric-reconstruction approaches often require large datasets, external localization, or complex modeling assumptions, reducing transparency and deployment suitability on resource-constrained platforms. We present an interpretable fuzzy-inference framework that generates continuous yaw commands from low-dimensional features extracted from YOLO boxes: target centroid location, area, and aspect ratio. No explicit geometric modeling is required. A Mamdani fuzzy system serves as an interpretable baseline using a shoulder--triangle--shoulder input partition. It is followed by a first-order Takagi--Sugeno model with three antecedent membership terms per input, whose parameters are derived from training-set quantiles, yielding a compact 27-rule structure. Evaluation uses 6{,}169 labeled samples from a VICON motion-capture environment. Across five randomized train--test splits, the Takagi--Sugeno model achieves a test-set mean absolute error of $0.140^\circ \pm 0.003^\circ$, a root mean squared error of $0.200^\circ \pm 0.008^\circ$, and a maximum absolute error of $1.254^\circ \pm 0.121^\circ$. Within-threshold accuracies are $99.676% \pm 0.270%$ for $\pm1^\circ$ and $100.000% \pm 0.000%$ for both $\pm3^\circ$ and $\pm5^\circ$. Directional consistency between image-plane horizontal displacement and predicted yaw sign reaches $90.254% \pm 0.612%$. These results show that the framework is transparent, data-efficient, computationally lightweight, and suitable for real-time vision-based UAV guidance toward mobile ground targets.

Many critical reasoning tasks, including clinical diagnosis, legal judgment, and industrial fault diagnosis, require step-dependent causal chains in which early errors propagate and correct conclusions can mask invalid reasoning. Although large language models perform well on such tasks, privacy, latency, and controllability motivate distillation into locally deployable models. Standard trajectory imitation does not correct process errors on the student's own rollout distribution. We propose CausalOPD, a curriculum online process distillation framework. A knowledge-augmented teacher first provides trajectories grounded in domain-specific causal rules, entity relations, and structural constraints. The student then generates on-policy trajectories, and the teacher identifies the first wrong step, defined as the earliest transition that verifiably violates available constraints. Starting from the verified prefix, short-horizon reinforcement learning repairs this localized failure. A causal-stage curriculum advances from evidence-level to mechanism-level and conclusion-level errors, following their propagation order. Across three domains, CausalOPD improves average path correctness by 23.4 percentage points over sequence-level online process distillation and reduces the right-label-wrong-reasoning rate from 15.7% to 4.4%. The domain-specific 8B students also surpass both evaluated proprietary references in path correctness across all domains.

Aligning diffusion models with human preferences usually relies on a sparse terminal reward evaluated on the final generated samples, presenting a severe temporal credit-assignment challenge across the multi-step denoising process. We propose Latent Reward Registers, a mechanism that estimates terminal preference directly from intermediate noisy latents by prepending learnable, position-free register tokens to the input sequence of a frozen Diffusion Transformer (DiT). This independent readout mechanism extracts latent reward evidence without altering the generator's hidden states or velocity field. The resulting dense, differentiable reward signal throughout the full denoising process facilitates two alignment strategies. For training, Reward-Gradient On-Policy Distillation (RG-OPD) distills reward-guided updates along on-policy trajectories, bypassing the computationally expensive rollouts of standard policy gradients. For inference, Reward-Guided Sampling (RGS) steers trajectories via magnitude-matched reward gradients without parameter updates. Empirically, at high noise levels (u = 0.8), the registers reach the highest pairwise accuracy among the evaluated latent reward models. Furthermore, RG-OPD outperforms online reinforcement learning baselines while reducing GPU hours by up to 33x, and RGS establishes a new state-of-the-art among training-free methods, strictly enhancing both alignment and perceptual metrics. Code and weights are available at https://github.com/Guanys-dar/latent-reward-register

We aim to improve model performance in multi-reward reinforcement learning training process. Existing Group reward-Decoupled Normalization Policy Optimization (GDPO) has mitigated the issue of reward signals masking one another during direct scalarization by normalizing each reward dimension separately before aggregation. However, our experiments show that GDPO still struggles to balance reward signals with different granularities. Specifically, in some particular training tasks, the model may receive a dense reward that assigns fine-grained scores ranging from 0.1 to 1.0, together with a sparse reward that provides only binary feedback of either 0 or 1. In such cases, we find that the sparse reward may provide an insufficient optimization signal, preventing its corresponding capability from being effectively reinforced. Therefore, how can we strengthen the optimization signal from the sparse reward without sacrificing the capability already learned from the fine-grained reward? To overcome this limitation, we propose Specialize-and-Merge Online Policy Distillation (SMOPD), a two-stage training method for multi-reward optimization. Stage1-Specialize: SMOPD first employs reward-priority configurations to train multiple reward-specialized teachers, allowing each reward to be learned under conditions where its signal can effectively drive optimization. Stage2-Merge: SMOPD then utilizes online policy distillation to combine the reward-specialized capabilities of these teachers into a single student policy, while maintaining balanced task-level optimization. To validate our method, we conduct experiments on two multi-reward settings: complementary rewards(tool-calling accuracy and format) and conflicting rewards (helpful and harmless rewards). Based on above settings, SMOPD outperforms GDPO across 1.5B, 3B and 7B backbones.

Federated learning enables collaborative model training across distributed edge devices while strictly preserving data privacy. To facilitate practical deployment on resource-constrained edge devices, Spiking Neural Networks (SNNs) have emerged as a promising alternative to traditional Artificial Neural Networks (ANNs) due to their sparse computing mechanisms and high energy efficiency. However, jointly training ANNs and SNNs exposes a challenge of representational misalignment, which is intrinsically caused by differences in information representation, specifically the semantic gap between continuous real-valued activations in ANNs and discrete spatio-temporal spikes in SNNs. To overcome this barrier, we propose AS-FedBridge, a novel federated learning framework tailored for mixed ANN-SNN clients. AS-FedBridge features a lightweight Bridge equipped with a Pseudo-Spike Interface, which effectively projects continuous signals into a spike-compatible space to facilitate ANN-SNN alignment. Given the absence of existing mixed ANN-SNN federated frameworks, we establish a comprehensive benchmark to evaluate against multiple advanced heterogeneous FL methods. Our empirical analysis demonstrates a positive correlation between the degree of ANN-SNN alignment and the collaborative FL performance. Across four datasets, AS-FedBridge consistently demonstrates advanced accuracy while mitigating extreme scale, architecture, and client heterogeneity challenge. Furthermore, our framework enables a highly controllable trade-off between model performance and resource efficiency. AS-FedBridge accomplishes these robust performance gains while introducing only marginal computational overhead, establishing a robust and practical foundation for mixed ANN-SNN federated learning systems.