We study post-training for dynamically routed mixture-of-experts (MoE) models, focusing on integrating independently optimized specialists while keeping consolidation efficient. Our approach combines calibrated multi-teacher on-policy distillation (C-MOPD), which matches each selected teacher on a local Top-$K$ support with teacher-wise loss normalization and clipping, with an execution path that uses post-capacity bounded-width dispatch, final-route replay, and asynchronous rollout--training overlap. On Turing-20B-A2B, the selected model achieves a four-domain normalized integration score of 0.9988 and retains at least 90.16\% of the measured specialist gain in every domain. At matched $K=32$, the full calibration objective scores 0.9962, compared with 0.9724 without calibration and 0.9834 without clipping. At a matched 8K response cap, the model exceeds both shared SFT and Qwen3-8B on all four reasoning-enabled mathematics and code benchmarks (unweighted mean: 85.92, versus 80.33 for Qwen3-8B and 74.21 for SFT), and improves over SFT on ten of twelve broader evaluation metrics. In a PPU-based profile, the combined execution configuration reduces consolidation step time from 398.7 to 139.3 seconds, an observed 2.86$\times$ speedup over serial execution. These results demonstrate strong specialist integration alongside more efficient consolidation, while reflecting the complete training and execution framework rather than any single component.
Wenlin Chen, Mingtian Zhang, Jiajun He, Zijing Ou, José Miguel Hernández-Lobato, Bernhard Schölkopf, David Barber
Score-based distillation methods train one-step diffusion models in two stages: they first train a teacher score model, then distill it into a one-step student model. However, this distillation process can introduce bias from two sources: errors in the teacher score model and student score estimate. We propose DiffRatio, a new approach for training one-step diffusion models without teacher supervision based on a density-ratio formulation that connects score-based distribution matching and GAN-style adversarial training. Instead of using a teacher score model to provide training targets, DiffRatio directly learns a log density ratio between the student and data distributions across diffusion time steps. As a pre-trained score model is only used to initialize the one-step generator and does not supervise training, this design simplifies the training pipeline, mitigates gradient estimation bias, and reduces the size of auxiliary networks. The learned density ratio can be used as a verifier, enabling a principled inference-time parallel scaling method that further improves sample quality without external rewards or extra sequential computation. DiffRatio achieves strong one-step generation results on CIFAR-10 and ImageNet (64x64 and 512x512), outperforming most teacher-supervised distillation methods.
Reinforcement learning (RL) has emerged as a paradigm for fine-tuning large-scale generative models, such as diffusion and flow models, to align with complex human preferences and user-specified tasks. A fundamental limitation remains \textit{the curse of diversity collapse}, where the objective formulation and optimization landscape inherently collapse the policy to a Dirac delta distribution. To address this challenge, we propose \textbf{DRIFT} (\textbf{D}ive\textbf{R}sity-\textbf{I}ncentivized Reinforcement \textbf{F}ine-\textbf{T}uning for Versatile Image Generation), an innovative framework that systematically incentivizes output diversity throughout the on-policy fine-tuning process, reconciling strong task alignment with high generation diversity to enhance versatility essential for applications that demand diverse candidate generations. We approach the problem across three representative perspectives: i) \textbf{sampling} a reward-concentrated subset that filters out reward outliers to prevent premature collapse; ii) \textbf{prompting} with stochastic variations to expand the conditioning space, and iii) \textbf{optimization} of the intra-group diversity with a potential-based reward shaping mechanism. Experimental results show that DRIFT exhibits clear Pareto dominance in task alignment and generation diversity, achieving 3.88\%$\sim$81.42\% higher diversity at matched alignment and 14.38\%$\sim$56.45\% higher alignment at matched diversity. Our code is available at \textcolor{magenta}{\href{https://github.com/NJU-RL/DRIFT}{https://github.com/NJU-RL/DRIFT}}.
Hanqing Wang, Zhaotian Cai, Ruichuan An, Bohan Zeng, Tao Zhang, Yilin Chen, Xu Jiang, Kaiquan Chen 等 15 位作者
Vision--Language--Action (VLA) models generate robot actions from visual observations and language instructions, but their understanding of physical dynamics remains limited. Video world models can capture scene dynamics, yet generating future videos introduces high latency into the control loop. We introduce PDRVLA, whose lightweight Physical Dynamics Module extracts physical tokens from vision--language features to condition action generation without online video generation. The module learns through direct prediction of future RGB, depth, and optical-flow features, with auxiliary supervision through a frozen pretrained video world model. Training relies entirely on publicly available data, expanded through human-to-robot synthesis, subtask annotation, and embodied chain-of-thought (ECoT) annotation. Across four simulation benchmarks, PDRVLA achieves success rates of 99.3% on LIBERO, 89.9% on LIBERO-Plus, 93.6% on RoboTwin~2.0, and 81.5% on SimplerEnv. Ablations show improved future prediction with world-model supervision and evaluate how subtask conditioning, future object-pose supervision, and ECoT training affect action success.
Long-term treatment effect estimation is a significant but challenging problem in many applications. Existing methods rely on ideal assumptions, such as no unobserved confounders or binary treatment, to estimate long-term average treatment effects. However, in numerous real-world applications, these assumptions could be violated, and average treatment effects are insufficient for personalized decision-making. In this paper, we address a more general problem of estimating long-term Heterogeneous Dose-Response Curve (HDRC) while accounting for unobserved confounders and continuous treatment. Specifically, to remove the unobserved confounders in the long-term observational data, we introduce an optimal transport weighting framework to align the long-term observational data to an auxiliary short-term experimental data. Furthermore, to accurately predict the heterogeneous effects of continuous treatment, we establish a generalization bound on counterfactual prediction error by leveraging the reweighted distribution induced by optimal transport. Finally, we develop a long-term HDRC estimator building upon the above theoretical foundations. Extensive experiments on synthetic and semi-synthetic datasets demonstrate the effectiveness of our approach.
Illicit account detection is a critical task for financial risk control systems, aiming to pinpoint malicious actors within transaction networks. Existing efforts typically recast transaction networks into directed multigraphs and leverage graph neural networks (GNNs) to detect anomalies via topological structures under the homophily assumption. However, directly grafting GNNs onto transaction graphs is faced with at least two pivotal challenges, i.e., structural inconsistency and label scarcity. The former arises from deceptive connections forged by malicious actors, while the latter stems from the lack of annotated data. In light of these, we introduce MINT, a novel graph neural framework that mitigates structural inconsistency via an attentive message-passing scheme on transaction graphs. Specifically, MINT introduces a discrepancy-aware feature propagation, which tracks malicious actors by quantifying deviations between nodes across bi-directional transaction flows. A transaction salience module is proposed to mitigate structural inconsistency by adaptively regularizing the interaction strength between nodes. Furthermore, to alleviate label scarcity, MINT is optimized via Positive-Unlabeled (PU) learning, which exploits only a minority of illicit nodes coupled with a majority of unlabeled ones. Extensive experiments conducted on three financial transaction benchmarks, including two large-scale public and one in-house real-world datasets, validate the superior effectiveness of MINT over several cutting-edge baselines.
Taisuke Yasuda, MohammadHossein Bateni, Lin Chen, Matthew Fahrbach, Gang Fu, Vahab Mirrokni
Feature selection is the problem of selecting a subset of features for a machine learning model that maximizes model quality subject to a budget constraint. For neural networks, prior methods, including those based on $\ell_1$ regularization, attention, and other techniques, typically select the entire feature subset in one evaluation round, ignoring the residual value of features during selection, i.e., the marginal contribution of a feature given that other features have already been selected. We propose a feature selection algorithm called Sequential Attention that achieves state-of-the-art empirical results for neural networks. This algorithm is based on an efficient one-pass implementation of greedy forward selection and uses attention weights at each step as a proxy for feature importance. We give theoretical insights into our algorithm for linear regression by showing that an adaptation to this setting is equivalent to the classical Orthogonal Matching Pursuit (OMP) algorithm, and thus inherits all of its provable guarantees. Our theoretical and empirical analyses offer new explanations towards the effectiveness of attention and its connections to overparameterization, which may be of independent interest.