Searcharxiv⌕ Search

arXiv subjects

Pengfei Zheng

Publications and source records attributed to Pengfei Zheng.

At least 19 recordsLinked to original sources

Exploration of Reinforcement Learning in Generative Recommendation Models

Traditional cascaded recommender systems balance accuracy and efficiency over large item spaces, but misaligned stage-wise objectives hinder end-to-end optimization. Generative recommendation unifies these stages in an autoregressive framework, providing a natural interface for reinforcement learning (RL) to optimize the entire generation process. However, directly applying RL often leads to unstable optimization and degraded recommendation quality. We revisit Group Relative Policy Optimization (GRPO) and other ratio-based objectives through a unified logit-gradient view. We identify three issues: (i) clipped surrogates create dead-gradient regions, under-utilizing samples; (ii) negative-advantage updates can become effectively unbounded as importance ratios grow, allowing a small subset of negatives to dominate optimization; and (iii) ratio-based control lacks sensitivity to absolute policy confidence, so updates on high-confidence cases may fail to vanish. Motivated by this diagnosis, we propose \emph{Gradient-Bounded Policy Optimization} (GBPO), a simple objective with stop-gradient normalization that is simultaneously (a) fully gradient-active, (b) sign-symmetrically bounded for both positive and negative updates, and (c) explicitly confidence-weighted under current policy probabilities. This ensures that updates gracefully vanish for $π\!\to\!1$ positives and $π\!\to\!0$ negatives. Across the evaluated settings, GBPO achieves competitive recommendation performance with more stable training than competing RL methods.

cs.LG↗

StrictPRMBench: A Replayable Evaluation Benchmark for Auditing Fixed-Score PRM Aggregation

Process reward models (PRMs) score reasoning steps, but best-of-N selection also requires a rule that aggregates those scores. We study that rule as a separate evaluation object. StrictPRMBench fixes mathematical reasoning pools, scorer evidence, and the source trace of a deletion intervention, so that different reducers face the same recorded selection problem. Across seven scoring checkpoints and two benchmarks, the stored N=32 outcomes give higher deletion-induced selection instability for product than geometric mean (GM) in all 14 cells. The equal-cell difference is +0.0965; a shared-problem bootstrap, conditional on these checkpoints and frozen pools, gives a 95% interval of [0.0846,0.1086]. A pre-injection-defined sensitivity excludes replacement slots that are either reducer's clean winner and retains 4,459 of 4,622 cell-problem records. The retained difference is +0.0986 [0.0875,0.1098]. Effects are heterogeneous: the two Qwen cells have positive percentile-bootstrap intervals on retained support but two-sided exact paired p-values of 0.0625 and 0.0703. Clean-pool Qwen accuracy does not detect a product advantage, but this is not an equivalence result. We provide versioned outcome records, CPU reanalysis, and an explicit evidence-availability contract. The contribution is a diagnostic evaluation protocol and a reproducible audit of recorded selector behavior, not a new PRM, a universally optimal reducer, or evidence of deployment-wide harm. Raw-score replay and outcome-record reanalysis are distinguished: missing primary edited-score vectors limit independent evaluation of new reducers under the same stress.

cs.LG↗

Evaluating the accuracy of KV cache reuse techniques

Position-independent KV cache reuse aims to reduce latency in retrieval-augmented generation by reusing chunk-level KV caches across prompts. We show that current evaluations of KV cache reuse techniques rely on measurements that fail to faithfully capture the loss of accuracy attributable to reuse, often artificially inflating the reported effectiveness. We also show that existing datasets do not exhibit the reuse dynamics needed to thoroughly evaluate such techniques. To address these issues, we propose an evaluation methodology that measures this accuracy loss without ambiguity and we introduce Boxoffice, a tool that programmatically generates evaluation datasets that exercise challenging KV cache reuse patterns.

cs.LG↗

Ascend-RaBitQ: Heterogeneous NPU-CPU Acceleration of Billion-Scale Similarity Search with 1-bit Quantization

Vector similarity search is a critical component of modern AI systems, but traditional CPU-based implementations face fundamental scalability bottlenecks for billion-scale corpora due to prohibitive computational overhead and memory bandwidth limitations. While Neural Processing Units (NPUs) offer orders-of-magnitude higher compute density, existing CPU/GPU-optimized 1-bit RaBitQ quantization implementations cannot be directly ported to NPU architectures due to fundamental hardware mismatches, and homogeneous design paradigms struggle to simultaneously balance accuracy, memory footprint, and performance. This paper presents Ascend-RaBitQ, the first heterogeneous NPU-CPU optimized IVF-RaBitQ system for billion-scale vector search, built on the core insight that decoupling coarse ranking (NPU) from fine ranking (CPU) allows each stage to leverage its optimal hardware, breaking the long-standing accuracy-memory-performance trade-off. We propose a three-stage heterogeneous execution path comprising AI Core-accelerated coarse ranking on 1-bit quantized vectors, on-device AI CPU Top-k processing, and host CPU fine re-ranking on full-precision vectors. We introduce four NPU architecture-native optimizations: fused AIC-AIV operators for parallel distance computation, computation flow restructuring to exploit rotation orthogonality, fine-grained index block-level load balancing that breaks query boundaries, and intra-NPU pipeline parallelism between AI Core and AI CPU to mask Top-k latency. Evaluation on standard datasets shows that Ascend-RaBitQ achieves 3.0X to 62.8X faster index construction than the CPU baseline, up to 11.7X throughput improvement over the fastest CPU IVF-RaBitQ implementation, and over two orders of magnitude over the mathematically equivalent CPU baseline, while demonstrating encouraging scalability on distributed multi-NPU systems.

cs.IR↗

OneReason Technical Report

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style ``think before answer'' paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

cs.IR↗

OmniGen2: Towards Instruction-Aligned Multimodal Generation

In this work, we introduce OmniGen2, a versatile and open-source generative model designed to provide a unified solution for diverse generation tasks, including text-to-image, image editing, and in-context generation. Unlike OmniGen v1, OmniGen2 features two distinct decoding pathways for text and image modalities, utilizing unshared parameters and a decoupled image tokenizer. This design enables OmniGen2 to build upon existing multimodal understanding models without the need to re-adapt VAE inputs, thereby preserving the original text generation capabilities. To facilitate the training of OmniGen2, we developed comprehensive data construction pipelines, encompassing image editing and in-context generation data. Additionally, we introduce a reflection mechanism tailored for image generation tasks and curate a dedicated reflection dataset based on OmniGen2. Despite its relatively modest parameter size, OmniGen2 achieves competitive results on multiple task benchmarks, including text-to-image and image editing. To further evaluate in-context generation, also referred to as subject-driven tasks, we introduce a new benchmark named OmniContext. OmniGen2 achieves state-of-the-art performance among open-source models in terms of consistency. We will release our models, training code, datasets, and data construction pipeline to support future research in this field. Project Page: https://vectorspacelab.github.io/OmniGen2; GitHub Link: https://github.com/VectorSpaceLab/OmniGen2

cs.CV↗

Semantic Parallelism: Redefining Efficient MoE Inference via Model-Data Co-Scheduling

Prevailing LLM serving engines employ expert parallelism (EP) to implement multi-device inference of massive MoE models. However, the efficiency of expert parallel inference is largely bounded by inter-device communication, as EP embraces expensive all-to-all collectives to route tokens to the remote experts if not collocating on the same GPU/NPU device. Nevertheless, state-of-the-art schemes treat expert device-placement and request (or token) device-scheduling as separate concerns, triggering excessive communication between them and compromising inference efficiency This paper proposes Semantic Parallelism, a novel parallelism paradigm that minimizes the steep communication costs in EP-centric MoE serving via model-data collaborative scheduling. We implement Semantic Parallelism in a framework called Sem-MoE. Sem-MoE maximally collocates experts and their activating tokens onto the same device using proactively modeled activation likelihood between them and introduces three key techniques: (1) Offline model scheduling, which preliminarily clusters and collocates experts onto devices based on their co-activation tendencies for certain classes of input. (2) Online inter-request data scheduling for Attention-DP setups, which proactively rebatches incoming requests onto the device that hosts experts most likely and frequently activated by the corresponding requests. (3) Online intra-request data scheduling for Attention-TP setups, which seamlessly fuses a token reshuffling procedure into the original inference pipeline and proactively reschedules tokens to devices to reduce dispersed remote routing. We build Sem-MoE into a prevailing LLM serving engine SGLANG. Experiments show our collaborative scheduling approach can effectively reduce the all-to-all communication volume in EP and achieve superior inference throughput compared to existing solutions.

cs.LG↗

OpenOneRec Technical Report

While the OneRec series has successfully unified the fragmented recommendation pipeline into an end-to-end generative framework, a significant gap remains between recommendation systems and general intelligence. Constrained by isolated data, they operate as domain specialists-proficient in pattern matching but lacking world knowledge, reasoning capabilities, and instruction following. This limitation is further compounded by the lack of a holistic benchmark to evaluate such integrated capabilities. To address this, our contributions are: 1) RecIF Bench & Open Data: We propose RecIF-Bench, a holistic benchmark covering 8 diverse tasks that thoroughly evaluate capabilities from fundamental prediction to complex reasoning. Concurrently, we release a massive training dataset comprising 96 million interactions from 160,000 users to facilitate reproducible research. 2) Framework & Scaling: To ensure full reproducibility, we open-source our comprehensive training pipeline, encompassing data processing, co-pretraining, and post-training. Leveraging this framework, we demonstrate that recommendation capabilities can scale predictably while mitigating catastrophic forgetting of general knowledge. 3) OneRec-Foundation: We release OneRec Foundation (1.7B and 8B), a family of models establishing new state-of-the-art (SOTA) results across all tasks in RecIF-Bench. Furthermore, when transferred to the Amazon benchmark, our models surpass the strongest baselines with an average 26.8% improvement in Recall@10 across 10 diverse datasets (Figure 1). This work marks a step towards building truly intelligent recommender systems. Nonetheless, realizing this vision presents significant technical and theoretical challenges, highlighting the need for broader research engagement in this promising direction.

cs.IR↗

OneRec-Think: In-Text Reasoning for Generative Recommendation

The powerful generative capacity of Large Language Models (LLMs) has instigated a paradigm shift in recommendation. However, existing generative models (e.g., OneRec) operate as implicit predictors, critically lacking the capacity for explicit and controllable reasoning-a key advantage of LLMs. To bridge this gap, we propose OneRec-Think, a unified framework that seamlessly integrates dialogue, reasoning, and personalized recommendation. OneRec-Think incorporates: (1) Itemic Alignment: cross-modal Item-Textual Alignment for semantic grounding; (2) Reasoning Activation: Reasoning Scaffolding to activate LLM reasoning within the recommendation context; and (3) Reasoning Enhancement, where we design a recommendation-specific reward function that accounts for the multi-validity nature of user preferences. Experiments across public benchmarks show state-of-the-art performance. Moreover, our proposed "Think-Ahead" architecture enables effective industrial deployment on Kuaishou, achieving a 0.159\% gain in APP Stay Time and validating the practical efficacy of the model's explicit reasoning capability.

cs.IR↗

OneRec-V2 Technical Report

Recent breakthroughs in generative AI have transformed recommender systems through end-to-end generation. OneRec reformulates recommendation as an autoregressive generation task, achieving high Model FLOPs Utilization. While OneRec-V1 has shown significant empirical success in real-world deployment, two critical challenges hinder its scalability and performance: (1) inefficient computational allocation where 97.66% of resources are consumed by sequence encoding rather than generation, and (2) limitations in reinforcement learning relying solely on reward models. To address these challenges, we propose OneRec-V2, featuring: (1) Lazy Decoder-Only Architecture: Eliminates encoder bottlenecks, reducing total computation by 94% and training resources by 90%, enabling successful scaling to 8B parameters. (2) Preference Alignment with Real-World User Interactions: Incorporates Duration-Aware Reward Shaping and Adaptive Ratio Clipping to better align with user preferences using real-world feedback. Extensive A/B tests on Kuaishou demonstrate OneRec-V2's effectiveness, improving App Stay Time by 0.467%/0.741% while balancing multi-objective recommendations. This work advances generative recommendation scalability and alignment with real-world feedback, representing a step forward in the development of end-to-end recommender systems.

cs.IR↗

CLEANet: Robust and Efficient Anomaly Detection in Contaminated Multivariate Time Series

Multivariate time series (MTS) anomaly detection is essential for maintaining the reliability of industrial systems, yet real-world deployment is hindered by two critical challenges: training data contamination (noises and hidden anomalies) and inefficient model inference. Existing unsupervised methods assume clean training data, but contamination distorts learned patterns and degrades detection accuracy. Meanwhile, complex deep models often overfit to contamination and suffer from high latency, limiting practical use. To address these challenges, we propose CLEANet, a robust and efficient anomaly detection framework in contaminated multivariate time series. CLEANet introduces a Contamination-Resilient Training Framework (CRTF) that mitigates the impact of corrupted samples through an adaptive reconstruction weighting strategy combined with clustering-guided contrastive learning, thereby enhancing robustness. To further avoid overfitting on contaminated data and improve computational efficiency, we design a lightweight conjugate MLP that disentangles temporal and cross-feature dependencies. Across five public datasets, CLEANet achieves up to 73.04% higher F1 and 81.28% lower runtime compared with ten state-of-the-art baselines. Furthermore, integrating CRTF into three advanced models yields an average 5.35% F1 gain, confirming its strong generalizability.

cs.LG↗

Centrum: Model-based Database Auto-tuning with Minimal Distributional Assumptions

Gaussian-Process-based Bayesian optimization (GP-BO), is a prevailing model-based framework for DBMS auto-tuning. However, recent work shows GP-BO-based DBMS auto-tuners significantly outperformed auto-tuners based on SMAC, which features random forest surrogate models; such results motivate us to rethink and investigate the limitations of GP-BO in auto-tuner design. We find the fundamental assumptions of GP-BO are widely violated when modeling and optimizing DBMS performance, while tree-ensemble-BOs (e.g., SMAC) can avoid the assumption pitfalls and deliver improved tuning efficiency and effectiveness. Moreover, we argue that existing tree-ensemble-BOs restrict further advancement in DBMS auto-tuning. First, existing tree-ensemble-BOs can only achieve distribution-free point estimates, but still impose unrealistic distributional assumptions on uncertainty estimates, compromising surrogate modeling and distort the acquisition function. Second, recent advances in gradient boosting, which can further enhance surrogate modeling against vanilla GP and random forest counterparts, have rarely been applied in optimizing DBMS auto-tuners. To address these issues, we propose a novel model-based DBMS auto-tuner, Centrum. Centrum improves distribution-free point and interval estimation in surrogate modeling with a two-phase learning procedure of stochastic gradient boosting ensembles. Moreover, Centrum adopts a generalized SGBE-estimated locally-adaptive conformal prediction to facilitate a distribution-free uncertainty estimation and acquisition function. To our knowledge, Centrum is the first auto-tuner to realize distribution-freeness, enhancing BO's practicality in DBMS auto-tuning, and the first to seamlessly fuse gradient boosting ensembles and conformal inference in BO. Extensive physical and simulation experiments on two DBMSs and three workloads show Centrum outperforms 21 SOTA methods.

cs.LG↗

Co-VisiON: Co-Visibility ReasONing on Sparse Image Sets of Indoor Scenes

Humans exhibit a remarkable ability to recognize co-visibility-the 3D regions simultaneously visible in multiple images-even when these images are sparsely distributed across a complex scene. This ability is foundational to 3D vision, robotic perception, and relies not only on low-level feature matching but also on high-level spatial reasoning and cognitive integration. Yet, it remains unclear whether current vision models can replicate this human-level proficiency. In this work, we introduce the Co-VisiON benchmark, designed to evaluate human-inspired co-visibility reasoning across more than 1,000 sparse-view indoor scenarios. Our results show that while co-visibility is often approached as a low-level feature-matching task, it remains challenging for existing vision models under sparse conditions. Notably, a proprietary vision-language model surpasses all vision-only baselines, but all models fall significantly short of human performance. This gap underscores the limitations of current architectures and motivates the need for models that integrate spatial and semantic information in a human-like manner. Inspired by human visual cognition, we propose a novel multi-view baseline, Covis, which achieves top performance among pure vision models and narrows the gap to the proprietary VLM. We hope our benchmark and findings will spur further advancements in developing vision models capable of robust, cognitively inspired reasoning in challenging, sparse environments. Our dataset and source code can be found at https://ai4ce.github.io/CoVISION.

cs.CV↗

Geometric-Facilitated Denoising Diffusion Model for 3D Molecule Generation

Denoising diffusion models have shown great potential in multiple research areas. Existing diffusion-based generative methods on de novo 3D molecule generation face two major challenges. Since majority heavy atoms in molecules allow connections to multiple atoms through single bonds, solely using pair-wise distance to model molecule geometries is insufficient. Therefore, the first one involves proposing an effective neural network as the denoising kernel that is capable to capture complex multi-body interatomic relationships and learn high-quality features. Due to the discrete nature of graphs, mainstream diffusion-based methods for molecules heavily rely on predefined rules and generate edges in an indirect manner. The second challenge involves accommodating molecule generation to diffusion and accurately predicting the existence of bonds. In our research, we view the iterative way of updating molecule conformations in diffusion process is consistent with molecular dynamics and introduce a novel molecule generation method named Geometric-Facilitated Molecular Diffusion (GFMDiff). For the first challenge, we introduce a Dual-Track Transformer Network (DTN) to fully excevate global spatial relationships and learn high quality representations which contribute to accurate predictions of features and geometries. As for the second challenge, we design Geometric-Facilitated Loss (GFLoss) which intervenes the formation of bonds during the training period, instead of directly embedding edges into the latent space. Comprehensive experiments on current benchmarks demonstrate the superiority of GFMDiff.

cs.LG↗

Significantly Enhanced Vacancy Diffusion in Mn-containing Alloys

Manipulating point defects for tailored macroscopic properties remains a formidable challenge in materials science. This study demonstrates a proof-of-principle for a universal law involving element Mn, significantly enhancing vacancy diffusion through an unprecedented anomalous Friedel Oscillations phenomenon, across most metals in the periodic table. The correlation between Mn-induced point-defect dynamic changes and intrinsic macro-properties is robustly validated through the first-principles theory and well-designed experiments. The physical origin stems from Mn's exceptionally large effective intra-elemental 3d electron interactions, surpassing the Coulomb attraction induced by vacancy and disrupting the electron screening effect. Given the ubiquitous nature of vacancies and their recognition as the most crucial defects influencing nearly all physical and mechanical properties of crystalline materials, this outcome may drive advances in a broad domain.

cond-mat.mtrl-sci↗

Technical Report for Argoverse Challenges on Unified Sensor-based Detection, Tracking, and Forecasting

This report presents our Le3DE2E solution for unified sensor-based detection, tracking, and forecasting in Argoverse Challenges at CVPR 2023 Workshop on Autonomous Driving (WAD). We propose a unified network that incorporates three tasks, including detection, tracking, and forecasting. This solution adopts a strong Bird's Eye View (BEV) encoder with spatial and temporal fusion and generates unified representations for multi-tasks. The solution was tested in the Argoverse 2 sensor dataset to evaluate the detection, tracking, and forecasting of 26 object categories. We achieved 1st place in Detection, Tracking, and Forecasting on the E2E Forecasting track in Argoverse Challenges at CVPR 2023 WAD.

cs.CV↗

Technical Report for Argoverse Challenges on 4D Occupancy Forecasting

This report presents our Le3DE2E_Occ solution for 4D Occupancy Forecasting in Argoverse Challenges at CVPR 2023 Workshop on Autonomous Driving (WAD). Our solution consists of a strong LiDAR-based Bird's Eye View (BEV) encoder with temporal fusion and a two-stage decoder, which combines a DETR head and a UNet decoder. The solution was tested on the Argoverse 2 sensor dataset to evaluate the occupancy state 3 seconds in the future. Our solution achieved 18% lower L1 Error (3.57) than the baseline and got the 1 place on the 4D Occupancy Forecasting task in Argoverse Challenges at CVPR 2023.

cs.CV↗

Mirage: Towards Low-interruption Services on Batch GPU Clusters with Reinforcement Learning

Accommodating long-running deep learning (DL) training and inference jobs is challenging on GPU clusters that use traditional batch schedulers, such as Slurm. Given fixed wall clock time limits, DL researchers usually need to run a sequence of batch jobs and experience long interruptions on overloaded machines. Such interruptions significantly lower the research productivity and QoS for services that are deployed in production. To mitigate the issues from interruption, we investigate a set of statistical learning and reinforcement learning (RL) techniques, including random forest, xgboost, Deep Q-Network, and policy gradient to design a proactive provisioner using production job traces from three GPU clusters. We follow the standard machine learning practice by partitioning each job trace into training and validation subsets, then train each model using the training subset and evaluate the generality using the validation subset. We introduce Mirage, a Slurm-compatible resource provisioner that integrates the candidate RL methods. Our experiments show that the Mirage can reduce the interruption by 17-100% and safeguard 23%-76% of jobs with zero interruption across varying load levels on the three clusters.

cs.DC↗