Zhengyu Zhang

Yuhang Zhang, Shishun Tian, Muxin Liao et al.

Video semantic segmentation (VSS) is beneficial for dealing with dynamic scenes due to the continuous property of the real-world environment. On the one hand, some methods alleviate the predicted inconsistent problem between continuous frames. On the other hand, other methods employ the previous frame as the prior information to assist in segmenting the current frame. Although the previous methods achieve superior performances on the independent and identically distributed (i.i.d) data, they can not generalize well on other unseen domains. Thus, we explore a new task, the video generalizable semantic segmentation (VGSS) task that considers both continuous frames and domain generalization. In this paper, we propose a class-wise non-salient region generalized (CNSG) framework for the VGSS task. Concretely, we first define the class-wise non-salient feature, which describes features of the class-wise non-salient region that carry more generalizable information. Then, we propose a class-wise non-salient feature reasoning strategy to select and enhance the most generalized channels adaptively. Finally, we propose an inter-frame non-salient centroid alignment loss to alleviate the predicted inconsistent problem in the VGSS task. We also extend our video-based framework to the image-based generalizable semantic segmentation (IGSS) task. Experiments demonstrate that our CNSG framework yields significant improvement in the VGSS and IGSS tasks.

Rui Zhang, Qiwei Wu, Zhengyu Zhang et al.

Large scale GPU-parallel reinforcement learning has changed what can be trained in robot simulation, yet most systems still optimize one specialist policy per task. We propose a construction methodology for turning structured manipulation task families into GPU-parallel multi-task RL benchmarks, and instantiate it as MT-Libero using LIBERO assets and task predicates in Isaac Lab. The resulting benchmark supports simultaneous reinforcement learning over heterogeneous task suites with parallel rendering, physics randomization, and state-input or visual-input policies. To make such training practical under sparse success signals and limited prior data, we further propose DGPO, an on-policy demonstration guided method that combines importance weighted PPO with adaptive behavior cloning on matched demonstration actions. DGPO enables a tunable preference toward demonstrated task distributions, outperforming both prior-free RL and existing demonstration-based methods while preserving the stability and online improvement benefits of on-policy PPO.

Patrick Yin, Tyler Westenbroek, Zhengyu Zhang et al.

Reinforcement learning in massively parallel physics simulations has driven major progress in sim-to-real robot learning. However, current approaches remain brittle and task-specific, relying on extensive per-task engineering to design rewards, curricula, and demonstrations. Even with this engineering, they often fail on long-horizon, contact-rich manipulation tasks and do not meaningfully scale with compute, as performance quickly saturates when training revisits the same narrow regions of state space. We introduce \Method, a simple and scalable framework that enables on-policy reinforcement learning to robustly solve a broad class of dexterous manipulation tasks using a single reward function, fixed algorithm hyperparameters, no curricula, and no human demonstrations. Our key insight is that long-horizon exploration can be dramatically simplified by using simulator resets to systematically expose the RL algorithm to the diverse set of robot-object interactions which underlie dexterous manipulation. \Method\ programmatically generates such resets with minimal human input, converting additional compute directly into broader behavioral coverage and continued performance gains. We show that \Method\ gracefully scales to long-horizon dexterous manipulation tasks beyond the capabilities of existing approaches and is able to learn robust policies over significantly wider ranges of initial conditions than baselines. Finally, we distill \Method \ into visuomotor policies which display robust retrying behavior and substantially higher success rates than baselines when transferred to the real world zero-shot. Project webpage: https://omnireset.github.io

Jie Peng, Zhang Cao, Huaizhi Qu et al.

Although Large Language Models (LLMs) have demonstrated remarkable capabilities, their massive parameter counts and associated extensive computing make LLMs' deployment the main part of carbon emission from nowadays AI applications. Compared to modern GPUs like H$100$, it would be significantly carbon-sustainable if we could leverage old-fashioned GPUs such as M$40$ (as shown in Figure 1, M$40$ only has one third carbon emission of H$100$'s) for LLM servings. However, the limited High Bandwidth Memory (HBM) available on such GPU often cannot support the loading of LLMs due to the gigantic model size and intermediate activation data, making their serving challenging. For instance, a LLaMA2 model with $70$B parameters typically requires $128$GB for inference, which substantially surpasses $24$GB HBM in a $3090$ GPU and remains infeasible even considering the additional $64$GB DRAM. To address this challenge, this paper proposes a mixed-precision with a model modularization algorithm to enable LLM inference on outdated hardware with resource constraints. (The precision denotes the numerical precision like FP16, INT8, INT4) and multi-level caching (M2Cache).) Specifically, our M2Cache first modulizes neurons in LLM and creates their importance ranking. Then, it adopts a dynamic sparse mixed-precision quantization mechanism in weight space to reduce computational demands and communication overhead at each decoding step. It collectively lowers the operational carbon emissions associated with LLM inference. Moreover, M2Cache introduces a three-level cache management system with HBM, DRAM, and SSDs that complements the dynamic sparse mixed-precision inference. To enhance communication efficiency, M2Cache maintains a neuron-level mixed-precision LRU cache in HBM, a larger layer-aware cache in DRAM, and a full model in SSD.

MohammadJavad Azizi, Sheldon M Ross, Zhengyu Zhang

We consider the problem of finding, through adaptive sampling, which of $n$ options (arms) has the largest mean. Our objective is to determine a rule which identifies the best arm with a fixed minimum confidence using as few observations as possible, i.e. this is a fixed-confidence (FC) best arm identification (BAI) in multi-armed bandits. We study such problems under the Bayesian setting with both Bernoulli and Gaussian arms. We propose to use the classical "vector at a time" (VT) rule, which samples each remaining arm once in each round. We show how VT can be implemented and analyzed in our Bayesian setting and be improved by early elimination. Our analysis show that these algorithms guarantee an optimal strategy under the prior. We also propose and analyze a variant of the classical "play the winner" (PW) algorithm. Numerical results show that these rules compare favorably with state-of-art algorithms.

Hanbin Dai, Liangbo Zhou, Feng Zhang et al.

In this paper, we focus on the coordinate representation in human pose estimation. While being the standard choice, heatmap based representation has not been systematically investigated. We found that the process of coordinate decoding (i.e. transforming the predicted heatmaps to the coordinates) is surprisingly significant for human pose estimation performance, which nevertheless was not recognised before. In light of the discovered importance, we further probe the design limitations of the standard coordinate decoding method and propose a principled distribution-aware decoding method. Meanwhile, we improve the standard coordinate encoding process (i.e. transforming ground-truth coordinates to heatmaps) by generating accurate heatmap distributions for unbiased model training. Taking them together, we formulate a novel Distribution-Aware coordinate Representation for Keypoint (DARK) method. Serving as a model-agnostic plug-in, DARK significantly improves the performance of a variety of state-of-the-art human pose estimation models. Extensive experiments show that DARK yields the best results on COCO keypoint detection challenge, validating the usefulness and effectiveness of our novel coordinate representation idea. The project page containing more details is at https://ilovepose.github.io/coco