arXiv subjects

Weichang Wu

Publications and source records attributed to Weichang Wu.

9 recordsLinked to original sources

arXiv2026

SkillCome: Group Contrast Skill Optimization with Dual Memory

Skill evolution improves the capabilities of large language models by analyzing trajectories generated under a given skill and modifying the skill accordingly. Existing approaches typically generate a single trajectory per question. However, this provides insufficient optimization signals since it requires inferring effective skill edits from a solitary path. It is difficult to pinpoint which actions caused the failure in a failed trajectory, or to determine which actions in a successful one should be incorporated into the skill. Furthermore, they rely on a local batch of trajectories for analysis, making the optimization direction susceptible to noisy evidence. To address these, we propose SkillCome, a Skill-evolution method based on group Contrast optimization with dual memory. For each question, SkillCome generates trajectories and performs group contrast analysis to precisely identify key behavioral divergences between successful and failed trajectories, offering reliable optimization signals. The dual memory system further accumulates evidence from historical steps to track patterns shared across different groups, leading to more generalized optimization directions. Together, SkillCome builds a systematic optimization process that transforms experience from observed successful trajectories into reusable skills. Extensive experiments on six benchmarks spanning question answering, reasoning, and agentic tasks demonstrate the effectiveness of our method. SkillCome consistently outperforms baselines across five models of varying families and scales, with gains up to +5.69 points.

cs.AI↗

arXiv2026

Towards Understanding On-Policy Distillation through the Lens of Test-Time Scaling

On-policy distillation (OPD) has emerged as a promising post-training technique for enhancing LLM reasoning. Under the reverse KL objective, the idealized optimum of OPD aligns the student distribution with that of the teacher. When the teacher consistently outperforms the student, this naturally suggests that OPD should yield broad improvements over the pre-OPD student. However, do such improvements extend across the entire range of test-time sampling budgets? In this work, we revisit this expectation through the lens of test-time scaling by varying the sampling budget $K$ and evaluating performance with pass@$K$. Across multiple settings, we observe two distinct patterns: OPD can improve pass@$K$ at both small and large sampling budgets, but it can also improve small-budget performance while reducing large-budget pass@$K$. We show one condition that guarantees such a reversal and an idealized reverse KL counterexample where it occurs even when the teacher has higher accuracy on every problem. To choose between two candidate teachers at a target sampling budget, we propose the \textit{Teacher Advantage Score at $K$} (TAS@$K$), which can be computed before OPD training to predict which teacher will lead to a larger improvement in pass@$K$. Across three domains and thirteen benchmarks, the ordering predicted by TAS@$K$ agrees with the observed pass@$K$ improvements of the resulting OPD models in 83.6\% of experiments, providing a useful signal for teacher selection at the target pass@$K$.

cs.LG↗

arXiv2026

Focal Reward: Balanced Reinforcement Learning under Rubric-Based Rewards

The open-ended generation in LLMs usually requires multi-dimensional rubrics to adequately assess quality and guide the improvement of reinforcement learning. However, a critical dilemma inherent in this training paradigm is the imbalanced reward polarization along different rubric dimensions. Under this bottleneck, even if LLMs achieve relatively high rewards after training, they may still exhibit severe deficiencies in certain dimensions, leading to a direct deterioration in user experience. To address this problem, we propose Focal Reward, a novel objective to automatically balance the training of reinforcement learning under rubric-based rewards. Specifically, we first leverage an inverse reward projection mechanism to estimate the saturation degree of each criterion in the rubric, which forms the basis to calibrate the reward direction. Then, the final objective is designed with an automatically reweighting coefficient for each criterion to achieve the fine-grained balancing. Extensive experiments across three model scales and six benchmarks demonstrate that our Focal Reward method outperforms the strongest static aggregation baseline in all 18 model-benchmark comparisons. Rollout, mechanism, and ablation analyses further show that these gains arise from online, saturation-aware reallocation toward rubrics that still have room for improvement.

cs.LG↗

arXiv2025

Bootstrapping your behavior: a new pretraining strategy for user behavior sequence data

User Behavior Sequence (UBS) modeling is crucial in industrial applications. As data scale and task diversity grow, UBS pretraining methods have become increasingly pivotal. State-of-the-art UBS pretraining methods rely on predicting behavior distributions. The key step in these methods is constructing a selected behavior vocabulary. However, this manual step is labor-intensive and prone to bias. The limitation of vocabulary capacity also directly affects models' generalization ability. In this paper, we introduce Bootstrapping Your Behavior (\model{}), a novel UBS pretraining strategy that predicts an automatically constructed supervision embedding summarizing all behaviors' information within a future time window, eliminating the manual behavior vocabulary selection. In implementation, we incorporate a student-teacher encoder scheme to construct the pretraining supervision effectively. Experiments on two real-world industrial datasets and eight downstream tasks demonstrate that \model{} achieves an average improvement of 3.9\% in AUC and 98.9\% in training throughput. Notably, the model exhibits meaningful attention patterns and cluster representations during pretraining without any label supervision. In our online deployment over two months, the pretrained model improves the KS by about 2.7\% and 7.1\% over the baseline model for two financial overdue risk prediction tasks in the Alipay mobile application, which reduces bad debt risk by millions of dollars for Ant group.

cs.LG↗

arXiv2024

An Adaptive Placement and Parallelism Framework for Accelerating RLHF Training

Recently, ChatGPT or InstructGPT like large language models (LLM) has made a significant impact in the AI world. Many works have attempted to reproduce the complex InstructGPT's training pipeline, namely Reinforcement Learning with Human Feedback (RLHF). However, the mainstream distributed RLHF training methods typically adopt a fixed model placement strategy, referred to as the Co-located strategy. This strategy treats all four interdependent models involved in RLHF as a single entity, distributing them across all devices and applying parallelism techniques designed for a single model, regardless of the workload heterogeneity inherent to each model. As a result, this strategy exacerbates the generation bottlenecks in the RLHF training and degrades the overall training efficiency. To address these issues, we propose a flexible model placement framework that offers two general and agile model placement strategies. The Interleaving strategy helps reduce memory redundancy and communication costs of RLHF training by placing models without dependencies on exclusive devices with careful orchestration. On the other hand, the Disaggregated strategy improves the throughput of model training by separating the training and inference runtime of the RLHF pipeline with additional shadow models. Furthermore, our framework provides a simple user interface and guidelines to easily and flexibly configure these strategies in various training scenarios. Our experiments have shown that our strategy can achieve notable improvements up to 11x, compared to the current state-of-the-art (SOTA) approaches. The results highlight the effectiveness and adaptability of our methods in accelerating the training of distributed RLHF.

cs.LG↗

arXiv2020

Modeling Event Propagation via Graph Biased Temporal Point Process

Temporal point process is widely used for sequential data modeling. In this paper, we focus on the problem of modeling sequential event propagation in graph, such as retweeting by social network users, news transmitting between websites, etc. Given a collection of event propagation sequences, conventional point process model consider only the event history, i.e. embed event history into a vector, not the latent graph structure. We propose a Graph Biased Temporal Point Process (GBTPP) leveraging the structural information from graph representation learning, where the direct influence between nodes and indirect influence from event history is modeled respectively. Moreover, the learned node embedding vector is also integrated into the embedded event history as side information. Experiments on a synthetic dataset and two real-world datasets show the efficacy of our model compared to conventional methods and state-of-the-art.

cs.SI↗

arXiv2019

Reinforcement Learning with Policy Mixture Model for Temporal Point Processes Clustering

Temporal point process is an expressive tool for modeling event sequences over time. In this paper, we take a reinforcement learning view whereby the observed sequences are assumed to be generated from a mixture of latent policies. The purpose is to cluster the sequences with different temporal patterns into the underlying policies while learning each of the policy model. The flexibility of our model lies in: i) all the components are networks including the policy network for modeling the intensity function of temporal point process; ii) to handle varying-length event sequences, we resort to inverse reinforcement learning by decomposing the observed sequence into states (RNN hidden embedding of history) and actions (time interval to next event) in order to learn the reward function, thus achieving better performance or increasing efficiency compared to existing methods using rewards over the entire sequence such as log-likelihood or Wasserstein distance. We adopt an expectation-maximization framework with the E-step estimating the cluster labels for each sequence, and the M-step aiming to learn the respective policy. Extensive experiments show the efficacy of our method against state-of-the-arts.

cs.LG↗

arXiv2018

Decoupled Learning for Factorial Marked Temporal Point Processes

This paper introduces the factorial marked temporal point process model and presents efficient learning methods. In conventional (multi-dimensional) marked temporal point process models, event is often encoded by a single discrete variable i.e. a marker. In this paper, we describe the factorial marked point processes whereby time-stamped event is factored into multiple markers. Accordingly the size of the infectivity matrix modeling the effect between pairwise markers is in power order w.r.t. the number of the discrete marker space. We propose a decoupled learning method with two learning procedures: i) directly solving the model based on two techniques: Alternating Direction Method of Multipliers and Fast Iterative Shrinkage-Thresholding Algorithm; ii) involving a reformulation that transforms the original problem into a Logistic Regression model for more efficient learning. Moreover, a sparse group regularizer is added to identify the key profile features and event labels. Empirical results on real world datasets demonstrate the efficiency of our decoupled and reformulated method. The source code is available online.

cs.LG↗

arXiv2016

Patient Flow Prediction via Discriminative Learning of Mutually-Correcting Processes

Over the past decade the rate of care unit (CU) use in the United States has been increasing. With an aging population and ever-growing demand for medical care, effective management of patients' transitions among different care facilities will prove indispensible for shortening the length of hospital stays, improving patient outcomes, allocating critical care resources, and reducing preventable re-admissions. In this paper, we focus on an important problem of predicting the so-called "patient flow" from longitudinal electronic health records (EHRs), which has not been explored via existing machine learning techniques. By treating a sequence of transition events as a point process, we develop a novel framework for modeling patient flow through various CUs and jointly predicting patients' destination CUs and duration days. Instead of learning a generative point process model via maximum likelihood estimation, we propose a novel discriminative learning algorithm aiming at improving the prediction of transition events in the case of sparse data. By parameterizing the proposed model as a mutually-correcting process, we formulate the estimation problem via generalized linear models, which lends itself to efficient learning based on alternating direction method of multipliers (ADMM). Furthermore, we achieve simultaneous feature selection and learning by adding a group-lasso regularizer to the ADMM algorithm. Additionally, for suppressing the negative influence of data imbalance on the learning of model, we synthesize auxiliary training data for the classes with extremely few samples, and improve the robustness of our learning method accordingly. Testing on real-world data, we show that our method obtains superior performance in terms of accuracy of predicting the destination CU transition and duration of each CU occupancy.

cs.LG↗