Jin Yang

Haiyang You, Chengwei Lou, Jin Yang

The rapid growth of hyperscale AI datacenters introduces structured, workload-driven active-power fluctuations at the point of interconnection. These fluctuations appear to the grid as time-varying disturbance injections that cannot be captured by conventional peak- or average-load representations. To reduce the residual power disturbance before it propagates into the bulk power system, this paper proposes a hybrid energy storage system with differentiable predictive control (HESS-DPC) framework for datacenter-side power smoothing. A workload-driven disturbance model is first established, representing the point-of-interconnection load deviation as the superposition of training and fine-tuning workloads to capture the structured forcing inputs that can excite generator frequency dynamics. A frequency-based rule-based controller then allocates this deviation between a battery energy storage system (BESS) and a supercapacitor (SC), assigning the energy-dominant component to the BESS and the fast-varying component to the SC. To overcome the anticipation and constraint limitations of fixed-frequency decomposition, a residual differentiable predictive control policy is trained offline to compute finite-horizon command corrections around the rule-based baseline while enforcing a one-step safeguard. Simulations on the NPCC 140-bus system show that HESS-DPC reduces grid-side residual deviations during workload transitions, improves SC state-of-charge sustainability over extended operation, and reduces generator peak-to-peak frequency deviations by more than 80 percent across all monitored generators, with the worst-affected generator response falling from 15.1 mHz to 1.3 mHz. These results confirm that local active-power smoothing at the datacenter point of interconnection can substantially mitigate frequency disturbances caused by AI workloads.

L. River Spencer, William D. Meador, Adrian Buganza Tepole et al.

Calibrating thermomechanical material models from experiments is challenging because deformation, temperature, and force responses are strongly coupled, while measurements are usually restricted to specimen surfaces. We present a full-field calibration framework for coupled finite-strain thermomechanical material models using boundary displacement, reaction-force data, and temperature. The forward model is formulated as a near-incompressible thermo-hyperelastic problem with thermomechanical coupling derived from a Helmholtz free energy, and the inverse problem is posed as a PDE-constrained optimization problem with weighted observation terms for the available data streams. Reduced gradients are computed with adjoint sensitivities that are obtained by automatic differentiation, enabling gradient-based calibration of nonlinear transient thermomechanical systems. The formulation is first verified on synthetic examples involving uniform thermal preconditioning and localized transient rod contact, where the ground-truth parameters are recovered from full-field measurements and force observations. The same workflow is then applied to experimental thermomechanical data by first calibrating a hyperelastic mechanical baseline from cyclic equibiaxial loading and subsequently identifying thermal expansion and directional shrinkage parameters from surface-temperature and boundary-force histories. The results demonstrate that coupled thermomechanical parameters can be inferred from experimentally accessible surface data without requiring volumetric observations.

Lei Liang, Mengshu Sun, Zhengke Gui et al.

The recently developed retrieval-augmented generation (RAG) technology has enabled the efficient construction of domain-specific applications. However, it also has limitations, including the gap between vector similarity and the relevance of knowledge reasoning, as well as insensitivity to knowledge logic, such as numerical values, temporal relations, expert rules, and others, which hinder the effectiveness of professional knowledge services. In this work, we introduce a professional domain knowledge service framework called Knowledge Augmented Generation (KAG). KAG is designed to address the aforementioned challenges with the motivation of making full use of the advantages of knowledge graph(KG) and vector retrieval, and to improve generation and reasoning performance by bidirectionally enhancing large language models (LLMs) and KGs through five key aspects: (1) LLM-friendly knowledge representation, (2) mutual-indexing between knowledge graphs and original chunks, (3) logical-form-guided hybrid reasoning engine, (4) knowledge alignment with semantic reasoning, and (5) model capability enhancement for KAG. We compared KAG with existing RAG methods in multihop question answering and found that it significantly outperforms state-of-theart methods, achieving a relative improvement of 19.6% on 2wiki and 33.5% on hotpotQA in terms of F1 score. We have successfully applied KAG to two professional knowledge Q&A tasks of Ant Group, including E-Government Q&A and E-Health Q&A, achieving significant improvement in professionalism compared to RAG methods.

Haiyang You, Chengwei Lou, Jin Zhao et al.

The expansion of data centers (DCs) drives a sustained increase in electricity demand and associated water withdrawals at generation sites. These withdrawals occur at generation sites and are virtually allocated to demand based on network power flows. Consequently, the actual water footprint of a specific load varies dynamically with generation dispatch and network conditions. Existing approaches typically rely on static statistical accounting to quantify these water footprints. However, such static methods fail to capture how dispatch optimization and workload relocation dynamically affect water withdrawals. As a result, static statistical accounting approaches remain decoupled from the optimization process, rendering them incapable of guiding workload relocation or power dispatch to mitigate water stress. To address this limitation, this paper develops an operational electricity-computation-water (ECW) nexus framework that internalizes virtual water impacts directly into power system dispatch. The framework represents dispatch optimization as a differentiable optimization layer embedded within a deep learning architecture, enabling efficient end-to-end learning of coordination policies while preserving operational feasibility. Combined with fixed-point coordination, the framework enforces consistency between virtual water attribution and physical generation-side withdrawals. Case studies on the IEEE 30-bus and 118-bus test systems demonstrate reliable convergence, exact power-water consistency, and reductions of approximately 3-5% in generation-related freshwater withdrawals under water-constrained conditions.

Zizheng Zhang, Yuyang Liao, Chen Chen et al.

Iterative code generation with Large Language Models (LLMs) can be viewed as an optimization process guided by textual feedback. However, existing LLM self-correction methods predominantly operate in a stateless, trial-and-error manner akin to first-order search, failing to leverage past problem-solving experiences. To bridge this gap, we introduce TextBFGS, a Case-Based Reasoning (CBR) framework inspired by the Quasi-Newton optimization method. Instead of retrieving raw, unstructured textual instances, TextBFGS maintains a dynamic Case Base of historical "Error-to-Operator" correction trajectories to approximate the semantic curvature (inverse Hessian matrix) of the task. Specifically, given a textual error feedback (the target problem), TextBFGS retrieves analogous historical correction patterns (Retrieve) and applies these abstract operators to refine the current code (Reuse/Revise). Furthermore, successful adaptations are continuously retained back into the Case Base (Retain), enabling a self-evolving system. Empirical evaluations on Python code optimization tasks (HumanEval, MBPP) demonstrate that TextBFGS significantly outperforms stateless baselines. It achieves superior pass rates with fewer model calls, establishing an efficient, experience-driven paradigm for LLM-based code optimization.

Zhu Han, Jin Yang, Lianru Gao et al.

Deep learning (DL) has been widely applied into hyperspectral image (HSI) classification owing to its promising feature learning and representation capabilities. However, limited by the spatial resolution of sensors, existing DL-based classification approaches mainly focus on pixel-level spectral and spatial information extraction through complex network architecture design, while ignoring the existence of mixed pixels in actual scenarios. To tackle this difficulty, we propose a novel dual-branch subpixel-guided network for HSI classification, called DSNet, which automatically integrates subpixel information and convolutional class features by introducing a deep autoencoder unmixing architecture to enhance classification performance. DSNet is capable of fully considering physically nonlinear properties within subpixels and adaptively generating diagnostic abundances in an unsupervised manner to achieve more reliable decision boundaries for class label distributions. The subpixel fusion module is designed to ensure high-quality information fusion across pixel and subpixel features, further promoting stable joint classification. Experimental results on three benchmark datasets demonstrate the effectiveness and superiority of DSNet compared with state-of-the-art DL-based HSI classification approaches. The codes will be available at https://github.com/hanzhu97702/DSNet, contributing to the remote sensing community.

Jiuyun Jiang, Yuecheng Hong, Bo Yang et al.

Modeling coordination among generative agents in complex multi-round decision-making presents a core challenge for AI and operations management. Although behavioral experiments have revealed cognitive biases behind supply chain inefficiencies, traditional methods face scalability and control limitations. We introduce a scalable experimental paradigm using Large Language Models (LLMs) to simulate multi-stage supply chain dynamics. Grounded in a Hierarchical Reasoning Framework, this study specifically analyzes the impact of cognitive heterogeneity on agent interactions. Unlike prior homogeneous settings, we employ DeepSeek and GPT agents to systematically vary reasoning sophistication across supply chain tiers. Through rigorously replicated and statistically validated simulations, we investigate how this cognitive diversity influences collective outcomes. Results indicate that agents exhibit myopic and self-interested behaviors that exacerbate systemic inefficiencies. However, we demonstrate that information sharing effectively mitigates these adverse effects. Our findings extend traditional behavioral methods and offer new insights into the dynamics of AI-enabled organizations. This work underscores both the potential and limitations of LLM-based agents as proxies for human decision-making in complex operational environments.

Boyong Wu, Chao Yan, Chen Hu et al.

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

Jin Yang, Peijie Qiu, Yichi Zhang et al.

Hierarchical transformers have achieved significant success in medical image segmentation due to their large receptive field and capabilities of effectively leveraging global long-range contextual information. Convolutional neural networks (CNNs) can also deliver a large receptive field by using large kernels, enabling them to achieve competitive performance with fewer model parameters. However, CNNs incorporated with large convolutional kernels remain constrained in adaptively capturing multi-scale features from organs with large variations in shape and size due to the employment of fixed-sized kernels. Additionally, they are unable to utilize global contextual information efficiently. To address these limitations, we propose Dynamic Large Kernel (DLK) and Dynamic Feature Fusion (DFF) modules. The DLK module employs multiple large kernels with varying kernel sizes and dilation rates to capture multi-scale features. Subsequently, a dynamic selection mechanism is utilized to adaptively highlight the most important spatial features based on global information. Additionally, the DFF module is proposed to adaptively fuse multi-scale local feature maps based on their global information. We integrate DLK and DFF in a hierarchical transformer architecture to develop a novel architecture, termed D-Net. D-Net is able to effectively utilize a multi-scale large receptive field and adaptively harness global contextual information. Extensive experimental results demonstrate that D-Net outperforms other state-of-the-art models in the two volumetric segmentation tasks, including abdominal multi-organ segmentation and multi-modality brain tumor segmentation. Our code is available at https://github.com/sotiraslab/DLK.

Jin Yang, Ping Wei, Huan Li et al.

Video moment retrieval and highlight detection are two highly valuable tasks in video understanding, but until recently they have been jointly studied. Although existing studies have made impressive advancement recently, they predominantly follow the data-driven bottom-up paradigm. Such paradigm overlooks task-specific and inter-task effects, resulting in poor model performance. In this paper, we propose a novel task-driven top-down framework TaskWeave for joint moment retrieval and highlight detection. The framework introduces a task-decoupled unit to capture task-specific and common representations. To investigate the interplay between the two tasks, we propose an inter-task feedback mechanism, which transforms the results of one task as guiding masks to assist the other task. Different from existing methods, we present a task-dependent joint loss function to optimize the model. Comprehensive experiments and in-depth ablation studies on QVHighlights, TVSum, and Charades-STA datasets corroborate the effectiveness and flexibility of the proposed framework. Codes are available at https://github.com/EdenGabriel/TaskWeave.

Weiwei Ma, Xiaobing Yu, Peijie Qiu et al.

In clinical practice, medical segmentation datasets are often limited and heterogeneous, with variations in modalities, protocols, and anatomical targets across institutions. Existing deep learning models struggle to jointly learn from such diverse data, often sacrificing either generalization or domain-specific knowledge. To overcome these challenges, we propose a joint training method called Universal Harmonization (U-Harmony), which can be integrated into deep learning-based architectures with a domain-gated head, enabling a single segmentation model to learn from heterogeneous datasets simultaneously. By integrating U-Harmony, our approach sequentially normalizes and then denormalizes feature distributions to mitigate domain-specific variations while preserving original dataset-specific knowledge. More appealingly, our framework also supports universal modality adaptation, allowing the seamless learning of new imaging modalities and anatomical classes. Extensive experiments on cross-institutional brain lesion datasets demonstrate the effectiveness of our approach, establishing a new benchmark for robust and adaptable 3D medical image segmentation models in real-world clinical settings.

Peijie Qiu, Jin Yang, Sayantan Kumar et al.

In the past decades, deep neural networks, particularly convolutional neural networks, have achieved state-of-the-art performance in a variety of medical image segmentation tasks. Recently, the introduction of the vision transformer (ViT) has significantly altered the landscape of deep segmentation models. There has been a growing focus on ViTs, driven by their excellent performance and scalability. However, we argue that the current design of the vision transformer-based UNet (ViT-UNet) segmentation models may not effectively handle the heterogeneous appearance (e.g., varying shapes and sizes) of objects of interest in medical image segmentation tasks. To tackle this challenge, we present a structured approach to introduce spatially dynamic components to the ViT-UNet. This adaptation enables the model to effectively capture features of target objects with diverse appearances. This is achieved by three main components: \textbf{(i)} deformable patch embedding; \textbf{(ii)} spatially dynamic multi-head attention; \textbf{(iii)} deformable positional encoding. These components were integrated into a novel architecture, termed AgileFormer. AgileFormer is a spatially agile ViT-UNet designed for medical image segmentation. Experiments in three segmentation tasks using publicly available datasets demonstrated the effectiveness of the proposed method. The code is available at \href{https://github.com/sotiraslab/AgileFormer}{https://github.com/sotiraslab/AgileFormer}.

Jin Yang, Xiaobing Yu, Peijie Qiu

Convolutional neural networks are widely used in various segmentation tasks in medical images. However, they are challenged to learn global features adaptively due to the inherent locality of convolutional operations. In contrast, MLP Mixers are proposed as a backbone to learn global information across channels with low complexity. However, they cannot capture spatial features efficiently. Additionally, they lack effective mechanisms to fuse and mix features adaptively. To tackle these limitations, we propose a novel Dynamic Decomposed Mixer module. It is designed to employ novel Mixers to extract features and aggregate information across different spatial locations and channels. Additionally, it employs novel dynamic mixing mechanisms to model inter-dependencies between channel and spatial feature representations and to fuse them adaptively. Subsequently, we incorporate it into a U-shaped Transformer-based architecture to generate a novel network, termed the Dynamic Decomposed MLP Mixer. We evaluated it for medical image segmentation on two datasets, and it achieved superior segmentation performance than other state-of-the-art methods.

Jin Yang, Daniel S. Marcus, Aristeidis Sotiras

U-Net has been widely used for segmenting abdominal organs, achieving promising performance. However, when it is used for multi-organ segmentation, first, it may be limited in exploiting global long-range contextual information due to the implementation of standard convolutions. Second, the use of spatial-wise downsampling (e.g., max pooling or strided convolutions) in the encoding path may lead to the loss of deformable or discriminative details. Third, features upsampled from the higher level are concatenated with those that persevered via skip connections. However, repeated downsampling and upsampling operations lead to misalignments between them and their concatenation degrades segmentation performance. To address these limitations, we propose Dynamically Calibrated Convolution (DCC), Dynamically Calibrated Downsampling (DCD), and Dynamically Calibrated Upsampling (DCU) modules, respectively. The DCC module can utilize global inter-dependencies between spatial and channel features to calibrate these features adaptively. The DCD module enables networks to adaptively preserve deformable or discriminative features during downsampling. The DCU module can dynamically align and calibrate upsampled features to eliminate misalignments before concatenations. We integrated the proposed modules into a standard U-Net, resulting in a new architecture, termed Dynamic U-Net. This architectural design enables U-Net to dynamically adjust features for different organs. We evaluated Dynamic U-Net in two abdominal multi-organ segmentation benchmarks. Dynamic U-Net achieved statistically improved segmentation accuracy compared with standard U-Net. Our code is available at https://github.com/sotiraslab/DynamicUNet.

Chengwei Lou, Zekai Jin, Wei Tang et al.

Real-time peer-to-peer (P2P) electricity markets dynamically adapt to fluctuations in renewable energy and variations in demand, maximizing economic benefits through instantaneous price responses while enhancing grid flexibility. However, scaling expert guidance for massive personalized prosumers poses critical challenges, including diverse decision-making demands and a lack of customized modeling frameworks. This paper proposes an integrated large language model-multi-agent reinforcement learning (LLM-MARL) framework for real-time P2P energy trading to address challenges such as the limited technical capability of prosumers, the lack of expert experience, and security issues of distribution networks. LLMs are introduced as experts to generate personalized strategies, guiding MARL under the centralized training with decentralized execution (CTDE) paradigm through imitation. To handle the scalability issues inherent in large-scale P2P networks, a differential attention-based critic network is introduced to efficiently extract key interaction features and enhance convergence. Experimental results demonstrate that LLM-generated strategies effectively substitute human experts. The proposed imitative expert MARL algorithms achieve significantly lower economic costs and voltage violation rates on test sets compared to baseline algorithms, while maintaining robust stability. This paper provides an effective solution for the real-time decision-making of the P2P electricity market by bridging expert knowledge with agent learning.

Jin Yang, Guangxin Jiang, Yinan Wang et al.

Recent years have witnessed exponential growth in developing deep learning (DL) models for time-series electricity forecasting in power systems. However, most of the proposed models are designed based on the designers' inherent knowledge and experience without elaborating on the suitability of the proposed neural architectures. Moreover, these models cannot be self-adjusted to dynamically changed data patterns due to the inflexible design of their structures. Although several recent studies have considered the application of the neural architecture search (NAS) technique for obtaining a network with an optimized structure in the electricity forecasting sector, their training process is computationally expensive and their search strategies are not flexible, indicating that the NAS application in this area is still at an infancy stage. In this study, we propose an intelligent automated architecture search (IAAS) framework for the development of time-series electricity forecasting models. The proposed framework contains three primary components, i.e., network function-preserving transformation operation, reinforcement learning (RL)-based network transformation control, and heuristic network screening, which aim to improve the search quality of a network structure. After conducting comprehensive experiments on two publicly-available electricity load datasets and two wind power datasets, we demonstrate that the proposed IAAS framework significantly outperforms the ten existing models or methods in terms of forecasting accuracy and stability. Finally, we perform an ablation experiment to showcase the importance of critical components in the proposed IAAS framework in improving forecasting accuracy.

Ling Team, Anqi Shen, Baihui Li et al.

We present Ring-1T, the first open-source, state-of-the-art thinking model with a trillion-scale parameter. It features 1 trillion total parameters and activates approximately 50 billion per token. Training such models at a trillion-parameter scale introduces unprecedented challenges, including train-inference misalignment, inefficiencies in rollout processing, and bottlenecks in the RL system. To address these, we pioneer three interconnected innovations: (1) IcePop stabilizes RL training via token-level discrepancy masking and clipping, resolving instability from training-inference mismatches; (2) C3PO++ improves resource utilization for long rollouts under a token budget by dynamically partitioning them, thereby obtaining high time efficiency; and (3) ASystem, a high-performance RL framework designed to overcome the systemic bottlenecks that impede trillion-parameter model training. Ring-1T delivers breakthrough results across critical benchmarks: 93.4 on AIME-2025, 86.72 on HMMT-2025, 2088 on CodeForces, and 55.94 on ARC-AGI-1. Notably, it attains a silver medal-level result on the IMO-2025, underscoring its exceptional reasoning capabilities. By releasing the complete 1T parameter MoE model to the community, we provide the research community with direct access to cutting-edge reasoning capabilities. This contribution marks a significant milestone in democratizing large-scale reasoning intelligence and establishes a new baseline for open-source model performance.

Zhenan Yu, Guangxin Jiang, Jin Yang

Recent studies on long-term time series forecasting have shown that simple linear models and MLP-based predictors can achieve strong performance without increasingly complex architectures. However, many competitive baselines still rely on structural priors such as frequency-domain modeling, explicit decomposition, multi-scale mixing, or sophisticated cross-variable interaction modules, while paying less attention to how simple temporal mappings should be trained and organized. In this paper, we propose STAIR, short for Stagewise Temporal Adaptation via Individualization and Residual Learning, a training paradigm for long-term time series forecasting that aims to unlock the capacity of simple temporal mapping models without introducing complex architectural modules. STAIR decomposes forecasting ability into three progressive stages: it first learns common temporal dynamics across variables through a shared temporal mapping, then adapts the shared model to each variable via channel-wise fine-tuning to capture variable-specific patterns, and finally complements the backbone with cross-variable information through residual learning. We further introduce Shared-to-Individual Fine-tuning and alpha-RevIN to mitigate the limitations of strict channel independence and the overly strong normalization prior induced by standard RevIN. This design gradually increases modeling flexibility while keeping the core temporal predictor as a shallow MLP in the main experiments, with linear variants analyzed separately. Experiments on nine long-term forecasting benchmarks show that STAIR matches or outperforms recent strong baselines while preserving a simple temporal backbone, providing a concise and effective modeling perspective for long-term time series forecasting.

Jin Yang, Ping Wei, Yixin Chen et al.

3D-aware visual pretraining has proven effective in improving the performance of downstream robotic manipulation tasks. However, existing methods are constrained to Euclidean embedding spaces, whose flat geometry limits their ability to model structural relations among embeddings. As a result, they struggle to learn structured embeddings that are essential for robust spatial perception in robotic applications. To this end, we propose HyperMVP, a self-supervised framework for \underline{Hyper}bolic \underline{M}ulti\underline{V}iew \underline{P}retraining. Hyperbolic space offers geometric properties well suited for capturing structural relations. Methodologically, we extend the masked autoencoder paradigm and design a GeoLink encoder to learn multiview hyperbolic representations. The pretrained encoder is then finetuned with visuomotor policies on manipulation tasks. In addition, we introduce 3D-MOV, a large-scale dataset comprising multiple types of 3D point clouds to support pretraining. We evaluate HyperMVP on COLOSSEUM, RLBench, and real-world scenarios, where it consistently outperforms strong baselines across diverse tasks and perturbation settings. Our results highlight the potential of 3D-aware pretraining in a non-Euclidean space for learning robust and generalizable robotic manipulation policies.

Jun Wang, Ninglun Gu, Kailai Zhang et al.

For Large Language Models (LLMs), a disconnect persists between benchmark performance and real-world utility. Current evaluation frameworks remain fragmented, prioritizing technical metrics while neglecting holistic assessment for deployment. This survey introduces an anthropomorphic evaluation paradigm through the lens of human intelligence, proposing a novel three-dimensional taxonomy: Intelligence Quotient (IQ)-General Intelligence for foundational capacity, Emotional Quotient (EQ)-Alignment Ability for value-based interactions, and Professional Quotient (PQ)-Professional Expertise for specialized proficiency. For practical value, we pioneer a Value-oriented Evaluation (VQ) framework assessing economic viability, social impact, ethical alignment, and environmental sustainability. Our modular architecture integrates six components with an implementation roadmap. Through analysis of 200+ benchmarks, we identify key challenges including dynamic assessment needs and interpretability gaps. It provides actionable guidance for developing LLMs that are technically proficient, contextually relevant, and ethically sound. We maintain a curated repository of open-source evaluation resources at: https://github.com/onejune2018/Awesome-LLM-Eval.

Tongxi Wu, Chenwei Xu, Jin Yang

The proliferation of cloud-integrated IoT systems has intensified exposure to Distributed Denial of Service (DDoS) attacks due to the expanded attack surface, heterogeneous device behaviors, and limited edge protection. However, DDoS detection in this context remains challenging because of complex traffic dynamics, severe class imbalance, and scarce labeled data. While recent methods have explored solutions to address class imbalance, many still struggle to generalize under limited supervision and dynamic traffic conditions. To overcome these challenges, we propose MixGAN, a hybrid detection method that integrates conditional generation, semi-supervised learning, and robust feature extraction. Specifically, to handle complex temporal traffic patterns, we design a 1-D WideResNet backbone composed of temporal convolutional layers with residual connections, which effectively capture local burst patterns in traffic sequences. To alleviate class imbalance and label scarcity, we use a pretrained CTGAN to generate synthetic minority-class (DDoS attack) samples that complement unlabeled data. Furthermore, to mitigate the effect of noisy pseudo-labels, we introduce a MixUp-Average-Sharpen (MAS) strategy that constructs smoothed and sharpened targets by averaging predictions over augmented views and reweighting them towards high-confidence classes. Experiments on NSL-KDD, BoT-IoT, and CICIoT2023 demonstrate that MixGAN achieves up to 2.5% higher accuracy and 4% improvement in both TPR and TNR compared to state-of-the-art methods, confirming its robustness in large-scale IoT-cloud environments. The source code is publicly available at https://github.com/0xCavaliers/MixGAN.

Jin Yang, Zhiqiang Wang, Yanbin Lin et al.

Recent advancements in large language models(LLMs), such as GPT-4 and GPT-4o, have shown exceptional performance, especially in languages with abundant resources like English, thanks to extensive datasets that ensure robust training. Conversely, these models exhibit limitations when processing under-resourced languages such as Chinese and Korean, where issues including hallucinatory responses remain prevalent. This paper traces the roots of these disparities to the tokenization process inherent to these models. Specifically, it explores how the tokenizers vocabulary, often used to speed up the tokenization process and reduce tokens but constructed independently of the actual model training data, inadequately represents non-English languages. This misrepresentation results in the propagation of under-trained or untrained tokens, which perpetuate biases and pose serious concerns related to data security and ethical standards. We aim to dissect the tokenization mechanics of GPT-4o, illustrating how its simplified token-handling methods amplify these risks and offer strategic solutions to mitigate associated security and ethical issues. Through this study, we emphasize the critical need to rethink tokenization frameworks to foster more equitable and secure AI technologies. The code and data are available at https://github.com/yeyimilk/LLMGPT4o

Suvadeep Banerjee, Steve Burns, Pasquale Cocchini et al.

This work focuses on an efficient Agile design methodology for domain-specific accelerators. We employ feature-by-feature enhancement of a vertical development stack and apply it to the TVM/VTA inference accelerator. We have enhanced the VTA design space and enabled end-to-end support for additional workloads. This has been accomplished by augmenting the VTA micro-architecture and instruction set architecture (ISA), as well as by enhancing the TVM compilation stack to support a wide range of VTA configs. The VTA tsim implementation (CHISEL-based) has been enhanced with fully pipelined versions of the ALU/GEMM execution units. In tsim, memory width can now range between 8-64 bytes. Field widths have been made more flexible to support larger scratchpads. New instructions have been added: element-wise 8-bit multiplication to support depthwise convolution, and load with a choice of pad values to support max pooling. Support for more layers and better double buffering has also been added. Fully pipelining ALU/GEMM helps significantly: 4.9x fewer cycles with minimal area change to run ResNet-18 under the default config. Configs featuring a further 11.5x decrease in cycle count at a cost of 12x greater area can be instantiated. Many points on the area-performance pareto curve are shown, showcasing the balance of execution unit sizing, memory interface width, and scratchpad sizing. Finally, VTA is now able to run Mobilenet 1.0 and all layers for ResNets, including the previously disabled pooling and fully connected layers. The TVM/VTA architecture has always featured end-to-end workload evaluation on RTL in minutes. With our modifications, it now offers a much greater number of feasible configurations with a wide range of cost vs. performance. All capabilities mentioned are available in opensource forks while a subset of these capabilities have already been upstreamed.

Josep Crego, Jungi Kim, Guillaume Klein et al.

Since the first online demonstration of Neural Machine Translation (NMT) by LISA, NMT development has recently moved from laboratory to production systems as demonstrated by several entities announcing roll-out of NMT engines to replace their existing technologies. NMT systems have a large number of training configurations and the training process of such systems is usually very long, often a few weeks, so role of experimentation is critical and important to share. In this work, we present our approach to production-ready systems simultaneously with release of online demonstrators covering a large variety of languages (12 languages, for 32 language pairs). We explore different practical choices: an efficient and evolutive open-source framework; data preparation; network architecture; additional implemented features; tuning for production; etc. We discuss about evaluation methodology, present our first findings and we finally outline further work. Our ultimate goal is to share our expertise to build competitive production systems for "generic" translation. We aim at contributing to set up a collaborative framework to speed-up adoption of the technology, foster further research efforts and enable the delivery and adoption to/by industry of use-case specific engines integrated in real production workflows. Mastering of the technology would allow us to build translation engines suited for particular needs, outperforming current simplest/uniform systems.

Yichi Zhang, Jin Yang, Yuchen Liu et al.

Semi-supervised learning has attracted much attention due to its less dependence on acquiring abundant annotations from experts compared to fully supervised methods, which is especially important for medical image segmentation which typically requires intensive pixel/voxel-wise labeling by domain experts. Although semi-supervised methods can improve the performance by utilizing unlabeled data, there are still gaps between fully supervised methods under extremely limited annotation scenarios. In this paper, we propose a simple yet efficient strategy to explore the usage of the Segment Anything Model (SAM) for enhancing semi-supervised medical image segmentation. Concretely, the segmentation model trained with domain knowledge provides information for localization and generating input prompts to the SAM. Then the generated pseudo-labels of SAM are utilized as additional supervision to assist in the learning procedure of the semi-supervised framework. Extensive experiments demonstrate that SemiSAM significantly improves the performance of existing semi-supervised frameworks when only one or a few labeled images are available and shows strong efficiency as a plug-and-play strategy for semi-supervised medical image segmentation.

Erdemt Bao, Jin Yang

In multi-agent collaborative sensing systems, substantial communication overhead from information exchange significantly limits scalability and real-time performance, especially in bandwidth-constrained environments. This often results in degraded performance and reduced reliability. To address this challenge, we propose WaveComm, a wavelet-based communication framework that drastically reduces transmission loads while preserving sensing performance in low-bandwidth scenarios. The core innovation of WaveComm lies in decomposing feature maps using Discrete Wavelet Transform (DWT), transmitting only compact low-frequency components to minimize communication overhead. High-frequency details are omitted, and their effects are reconstructed at the receiver side using a lightweight generator. A Multi-Scale Distillation (MSD) Loss is employed to optimize the reconstruction quality across pixel, structural, semantic, and distributional levels. Experiments on the OPV2V and DAIR-V2X datasets for LiDAR-based and camera-based perception tasks demonstrate that WaveComm maintains state-of-the-art performance even when the communication volume is reduced to 86.3% and 87.0% of the original, respectively. Compared to existing approaches, WaveComm achieves competitive improvements in both communication efficiency and perception accuracy. Ablation studies further validate the effectiveness of its key components.

Hao Gui, Lin Hu, Rui Chen et al.

3D Gaussian Splatting (3DGS) is increasingly attracting attention in both academia and industry owing to its superior visual quality and rendering speed. However, training a 3DGS model remains a time-intensive task, especially in load imbalance scenarios where workload diversity among pixels and Gaussian spheres causes poor renderCUDA kernel performance. We introduce Balanced 3DGS, a Gaussian-wise parallelism rendering with fine-grained tiling approach in 3DGS training process, perfectly solving load-imbalance issues. First, we innovatively introduce the inter-block dynamic workload distribution technique to map workloads to Streaming Multiprocessor(SM) resources within a single GPU dynamically, which constitutes the foundation of load balancing. Second, we are the first to propose the Gaussian-wise parallel rendering technique to significantly reduce workload divergence inside a warp, which serves as a critical component in addressing load imbalance. Based on the above two methods, we further creatively put forward the fine-grained combined load balancing technique to uniformly distribute workload across all SMs, which boosts the forward renderCUDA kernel performance by up to 7.52x. Besides, we present a self-adaptive render kernel selection strategy during the 3DGS training process based on different load-balance situations, which effectively improves training efficiency.

Tianxing Chen, Kaixuan Wang, Zhaohui Yang et al.

Embodied Artificial Intelligence (Embodied AI) is an emerging frontier in robotics, driven by the need for autonomous systems that can perceive, reason, and act in complex physical environments. While single-arm systems have shown strong task performance, collaborative dual-arm systems are essential for handling more intricate tasks involving rigid, deformable, and tactile-sensitive objects. To advance this goal, we launched the RoboTwin Dual-Arm Collaboration Challenge at the 2nd MEIS Workshop, CVPR 2025. Built on the RoboTwin Simulation platform (1.0 and 2.0) and the AgileX COBOT-Magic Robot platform, the competition consisted of three stages: Simulation Round 1, Simulation Round 2, and a final Real-World Round. Participants totally tackled 17 dual-arm manipulation tasks, covering rigid, deformable, and tactile-based scenarios. The challenge attracted 64 global teams and over 400 participants, producing top-performing solutions like SEM and AnchorDP3 and generating valuable insights into generalizable bimanual policy learning. This report outlines the competition setup, task design, evaluation methodology, key findings and future direction, aiming to support future research on robust and generalizable bimanual manipulation policies. The Challenge Webpage is available at https://robotwin-benchmark.github.io/cvpr-2025-challenge/.

Xiaojun Wu, Xiaoguang Jiang, Huiyang Li et al.

Large language models (LLMs) demonstrate remarkable reasoning capabilities in tasks such as algorithmic coding and mathematical problem-solving. Recent methods have improved reasoning through expanded corpus and multistage training combining reinforcement learning and supervised fine-tuning. Although some methods suggest that small but targeted dataset can incentivize reasoning via only distillation, a reasoning scaling laws is still taking shape, increasing computational costs. To address this, we propose a data-efficient distillation framework (DED) that optimizes the Pareto frontier of reasoning distillation. Inspired by the on-policy learning and diverse roll-out strategies of reinforcement learning, the key idea of our approach is threefold: (1) We identify that benchmark scores alone do not determine an effective teacher model. Through comprehensive comparisons of leading reasoning LLMs, we develop a method to select an optimal teacher model. (2) While scaling distillation can enhance reasoning, it often degrades out-of-domain performance. A carefully curated, smaller corpus achieves a balanced trade-off between in-domain and out-of-domain capabilities. (3) Diverse reasoning trajectories encourage the student model to develop robust reasoning skills. We validate our method through evaluations on mathematical reasoning (AIME 2024/2025, MATH-500) and code generation (LiveCodeBench), achieving state-of-the-art results with only 0.8k carefully curated examples, bypassing the need for extensive scaling. Our systematic analysis demonstrates that DED outperforms existing methods by considering factors beyond superficial hardness, token length, or teacher model capability. This work offers a practical and efficient pathway to advanced reasoning while preserving general capabilities.

Yuyan Shi, Jialu Ma, Jin Yang et al.

Medical image segmentation plays an important role in many image-guided clinical approaches. However, existing segmentation algorithms mostly rely on the availability of fully annotated images with pixel-wise annotations for training, which can be both labor-intensive and expertise-demanding, especially in the medical imaging domain where only experts can provide reliable and accurate annotations. To alleviate this challenge, there has been a growing focus on developing segmentation methods that can train deep models with weak annotations, such as image-level, bounding boxes, scribbles, and points. The emergence of vision foundation models, notably the Segment Anything Model (SAM), has introduced innovative capabilities for segmentation tasks using weak annotations for promptable segmentation enabled by large-scale pre-training. Adopting foundation models together with traditional learning methods has increasingly gained recent interest research community and shown potential for real-world applications. In this paper, we present a comprehensive survey of recent progress on annotation-efficient learning for medical image segmentation utilizing weak annotations before and in the era of foundation models. Furthermore, we analyze and discuss several challenges of existing approaches, which we believe will provide valuable guidance for shaping the trajectory of foundational models to further advance the field of medical image segmentation.

Xiaobing Yu, Jin Yang, Xiao Wu et al.

How to adapt a pre-trained model continuously for sequential tasks with different prediction class labels and domains and finally learn a generalizable model across diverse tasks is a long-lasting challenge. Continual learning (CL) has emerged as a promising approach to leverage pre-trained models (e.g., Transformers) for sequential tasks. While many existing CL methods incrementally store additional learned structures, such as Low-Rank Adaptation (LoRA) adapters or prompts and sometimes even preserve features from previous samples to maintain performance. This leads to unsustainable parameter growth and escalating storage costs as the number of tasks increases. Moreover, current approaches often lack task similarity awareness, which further hinders the models ability to effectively adapt to new tasks without interfering with previously acquired knowledge. To address these challenges, we propose FM-LoRA, a novel and efficient low-rank adaptation method that integrates both a dynamic rank selector (DRS) and dynamic meta-prompting (DMP). This framework allocates model capacity more effectively across tasks by leveraging a shared low-rank subspace critical for preserving knowledge, thereby avoiding continual parameter expansion. Extensive experiments on various CL benchmarks, including ImageNet-R, CIFAR100, and CUB200 for class-incremental learning (CIL), and DomainNet for domain-incremental learning (DIL), with Transformers backbone demonstrate that FM-LoRA effectively mitigates catastrophic forgetting while delivering robust performance across a diverse range of tasks and domains.

Peijie Qiu, Wenhui Zhu, Sayantan Kumar et al.

Multiple signal modalities, such as vision and sounds, are naturally present in real-world phenomena. Recently, there has been growing interest in learning generative models, in particular variational autoencoder (VAE), to for multimodal representation learning especially in the case of missing modalities. The primary goal of these models is to learn a modality-invariant and modality-specific representation that characterizes information across multiple modalities. Previous attempts at multimodal VAEs approach this mainly through the lens of experts, aggregating unimodal inference distributions with a product of experts (PoE), a mixture of experts (MoE), or a combination of both. In this paper, we provide an alternative generic and theoretical formulation of multimodal VAE through the lens of barycenter. We first show that PoE and MoE are specific instances of barycenters, derived by minimizing the asymmetric weighted KL divergence to unimodal inference distributions. Our novel formulation extends these two barycenters to a more flexible choice by considering different types of divergences. In particular, we explore the Wasserstein barycenter defined by the 2-Wasserstein distance, which better preserves the geometry of unimodal distributions by capturing both modality-specific and modality-invariant representations compared to KL divergence. Empirical studies on three multimodal benchmarks demonstrated the effectiveness of the proposed method.

Jin Yang, Qiong Wu, Zhiying Feng et al.

Large Language Models (LLMs) have demonstrated remarkable capabilities, leading to a significant increase in user demand for LLM services. However, cloud-based LLM services often suffer from high latency, unstable responsiveness, and privacy concerns. Therefore, multiple LLMs are usually deployed at the network edge to boost real-time responsiveness and protect data privacy, particularly for many emerging smart mobile and IoT applications. Given the varying response quality and latency of LLM services, a critical issue is how to route user requests from mobile and IoT devices to an appropriate LLM service (i.e., edge LLM expert) to ensure acceptable quality-of-service (QoS). Existing routing algorithms fail to simultaneously address the heterogeneity of LLM services, the interference among requests, and the dynamic workloads necessary for maintaining long-term stable QoS. To meet these challenges, in this paper we propose a novel deep reinforcement learning (DRL)-based QoS-aware LLM routing framework for sustained high-quality LLM services. Due to the dynamic nature of the global state, we propose a dynamic state abstraction technique to compactly represent global state features with a heterogeneous graph attention network (HAN). Additionally, we introduce an action impact estimator and a tailored reward function to guide the DRL agent in maximizing QoS and preventing latency violations. Extensive experiments on both Poisson and real-world workloads demonstrate that our proposed algorithm significantly improves average QoS and computing resource efficiency compared to existing baselines.

Jin Yang, Daniel S. Marcus, Aristeidis Sotiras

Medical Vision Foundation Models (Med-VFMs) have superior capabilities of interpreting medical images due to the knowledge learned from self-supervised pre-training with extensive unannotated images. To improve their performance on adaptive downstream evaluations, especially segmentation, a few samples from target domains are selected randomly for fine-tuning them. However, there lacks works to explore the way of adapting Med-VFMs to achieve the optimal performance on target domains efficiently. Thus, it is highly demanded to design an efficient way of fine-tuning Med-VFMs by selecting informative samples to maximize their adaptation performance on target domains. To achieve this, we propose an Active Source-Free Domain Adaptation (ASFDA) method to efficiently adapt Med-VFMs to target domains for volumetric medical image segmentation. This ASFDA employs a novel Active Learning (AL) method to select the most informative samples from target domains for fine-tuning Med-VFMs without the access to source pre-training samples, thus maximizing their performance with the minimal selection budget. In this AL method, we design an Active Test Time Sample Query strategy to select samples from the target domains via two query metrics, including Diversified Knowledge Divergence (DKD) and Anatomical Segmentation Difficulty (ASD). DKD is designed to measure the source-target knowledge gap and intra-domain diversity. It utilizes the knowledge of pre-training to guide the querying of source-dissimilar and semantic-diverse samples from the target domains. ASD is designed to evaluate the difficulty in segmentation of anatomical structures by measuring predictive entropy from foreground regions adaptively. Additionally, our ASFDA method employs a Selective Semi-supervised Fine-tuning to improve the performance and efficiency of fine-tuning by identifying samples with high reliability from unqueried ones.

Letian Kang, Shixian Luo, Yiqiang Li et al.

Vision-Language Models (VLMs) have demonstrated strong capabilities in multimodal understanding and generation tasks. However, their application to long video understanding remains hindered by the quadratic complexity of standard attention mechanisms. In this work, we introduce \textbf{PEVLM}, a fine-tuning-free parallel encoding method designed to enhance the prefilling efficiency of VLMs in long video scenarios. PEVLM partitions the input video into context blocks with a shared sink block, while preserving sequential position embeddings to align the attention weight distribution with that of Full-Attention. This design reduces attention complexity from $O((T \times N)^2)$ to $O(T \times N)$ where $T$ is the number of frames and $N$ the number of tokens per frame, without sacrificing accuracy. Extensive experiments across multiple state-of-the-art models and benchmarks demonstrate that PEVLM consistently outperforms existing parallel encoding approaches, achieving up to \textbf{7.47x} speedup in attention computation and reducing end-to-end latency by \textbf{40\%}. Remarkably, PEVLM not only maintains high accuracy, but in some settings even surpasses Full-Attention performance. Under strict latency constraints, it achieves substantial gains, improving accuracy from \textbf{23.26\%} to \textbf{61.03\%}. These results underscore the effectiveness of PEVLM for low-latency, long-context video understanding, making it a promising solution for real-world applications.

Lehu Bu, Zhaohan Yu, Shaoting Lin et al.

Laser-induced inertial cavitation (LIC)-where microscale vapor bubbles nucleate due to a focused high-energy pulsed laser and then violently collapse under surrounding high local pressures-offers a unique opportunity to investigate soft biological material mechanics at extremely high strain rates (>1000 1/s). Traditional rheological tools are often limited in these regimes by loading speed, resolution, or invasiveness. Here we introduce novel machine learning (ML) based microrheological frameworks that leverage LIC to characterize the viscoelastic properties of biological materials at ultra-high strain rates. We utilize ultra-high-speed imaging to capture time-resolved bubble radius dynamics during LIC events in various soft viscoelastic materials. These bubble radius versus time measurements are then analyzed using a newly developed Bubble Dynamics Transformer (BDT), a neural network trained on physics-based simulation data. The BDT accurately infers material viscoelastic parameters, eliminating the need for iterative fitting or complex inversion processes. This enables fast, accurate, and non-contact characterization of soft materials under extreme loading conditions, with significant implications for biomedical applications and materials science.

Hao Zheng, Md Rubel Ahmed, Parijat Mukherjee et al.

Concise and abstract models of system-level behaviors are invaluable in design analysis, testing, and validation. In this paper, we consider the problem of inferring models from communication traces of system-on-chip~(SoC) designs. The traces capture communications among different blocks of a SoC design in terms of messages exchanged. The extracted models characterize the system-level communication protocols governing how blocks exchange messages, and coordinate with each other to realize various system functions. In this paper, the above problem is formulated as a constraint satisfaction problem, which is then fed to a SMT solver. The solutions returned by the SMT solver are used to extract the models that accept the input traces. In the experiments, we demonstrate the proposed approach with traces collected from a transaction-level simulation model of a multicore SoC design and traces of a more detailed multicore SoC design developed in GEM5 environment.

Gang Peng, Jin Yang, Xinde Lia et al.

(This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.) To improve the efficiency of deep reinforcement learning (DRL)-based methods for robot manipulator trajectory planning in random working environments, we present three dense reward functions. These rewards differ from the traditional sparse reward. First, a posture reward function is proposed to speed up the learning process with a more reasonable trajectory by modeling the distance and direction constraints, which can reduce the blindness of exploration. Second, a stride reward function is proposed to improve the stability of the learning process by modeling the distance and movement distance of joint constraints. Finally, in order to further improve learning efficiency, we are inspired by the cognitive process of human behavior and propose a stage incentive mechanism, including a hard stage incentive reward function and a soft stage incentive reward function. Extensive experiments show that the soft stage incentive reward function is able to improve the convergence rate by up to 46.9% with the state-of-the-art DRL methods. The percentage increase in the convergence mean reward was 4.4-15.5% and the percentage decreases with respect to standard deviation were 21.9-63.2%. In the evaluation experiments, the success rate of trajectory planning for a robot manipulator reached 99.6%.

Jiancheng Qin, Jin Yang, Ying Chen et al.

Two-stage ensemble-based forecasting methods have been studied extensively in the wind power forecasting field. However, deep learning-based wind power forecasting studies have not investigated two aspects. In the first stage, different learning structures considering multiple inputs and multiple outputs have not been discussed. In the second stage, the model extrapolation issue has not been investigated. Therefore, we develop four deep neural networks for the first stage to learn data features considering the input-and-output structure. We then explore the model extrapolation issue in the second stage using different modeling methods. Considering the overfitting issue, we propose a new moving window-based algorithm using a validation set in the first stage to update the training data in both stages with two different moving window processes.Experiments were conducted at three wind farms, and the results demonstrate that the model with single input multiple output structure obtains better forecasting accuracy compared to existing models. In addition, the ridge regression method results in a better ensemble model that can further improve forecasting accuracy compared to existing machine learning methods. Finally, the proposed two-stage forecasting algorithm can generate more accurate and stable results than existing algorithms.

MD Rubel Ahmed, Hao Zheng, Parijat Mukherjee et al.

Comprehensive and well-defined specifications are necessary to perform rigorous and thorough validation of system-on-chip (SoC) designs. Message flows specify how components of an SoC design communicate and coordinate with each other to realize various system functions. Message flow specifications are essential for efficient system-level validation and debug for SoC designs. However, in practice such specifications are usually not available, often ambiguous, incomplete, or even contain errors. This paper addresses that problem by proposing a specification mining framework, FlowMiner, that automatically extracts message flows from SoC execution traces, which, unlike software traces, show a high degree of concurrency. A set of inference rules and optimization techniques are presented to improve mining performance and reduce mining complexity. Evaluation of this framework in several experiments shows promising results.

Yuting Cao, Parijat Mukherjee, Mahesh Ketkar et al.

Comprehensive specifications are essential for various activities across the entire validation continuum for system-on-chip (SoC) designs. However, specifications are often ambiguous, incomplete, or even contain inconsistencies or errors. This paper addresses this problem by developing a specification mining approach that automatically extracts sequential patterns from SoC transaction-level traces such that the mined patterns collectively characterize system-level specifications for SoC designs. This approach exploits long short-term memory (LSTM) networks trained with the collected SoC execution traces to capture sequential dependencies among various communication events. Then, a novel algorithm is developed to efficiently extract sequential patterns on system-level communications from the trained LSTM models. Several trace processing techniques are also proposed to enhance the mining performance. We evaluate the proposed approach on simulation traces of a non-trivial multi-core SoC prototype. Initial results show that the proposed approach is capable of extracting various patterns on system-level specifications from the highly concurrent SoC execution traces.

Gang Peng, Yuezhi Zheng, Jianfeng Li et al.

At present, most high-accuracy single-person pose estimation methods have high computational complexity and insufficient real-time performance due to the complex structure of the network model. However, a single-person pose estimation method with high real-time performance also needs to improve its accuracy due to the simple structure of the network model. It is currently difficult to achieve both high accuracy and real-time performance in single-person pose estimation. For use in human-machine cooperative operations, this paper proposes a single-person upper limb pose estimation method based on an end-to-end approach for accurate and real-time limb pose estimation. Using the stacked hourglass network model, a single-person upper limb skeleton key point detection model was designed.Deconvolution was employed to replace the up-sampling operation of the hourglass module in the original model, solving the problem of rough feature maps. Integral regression was used to calculate the position coordinates of key points of the skeleton, reducing quantization errors and calculations. Experiments showed that the developed single-person upper limb skeleton key point detection model achieves high accuracy and that the pose estimation method based on the end-to-end approach provides high accuracy and real-time performance.

Yongxi Tan, Jin Yang, Xin Chen et al.

Mobile network that millions of people use every day is one of the most complex systems in the world. Optimization of mobile network to meet exploding customer demand and reduce capital/operation expenditures poses great challenges. Despite recent progress, application of deep reinforcement learning (DRL) to complex real world problem still remains unsolved, given data scarcity, partial observability, risk and complex rules/dynamics in real world, as well as the huge reality gap between simulation and real world. To bridge the reality gap, we introduce a Sim-to-Real framework to directly transfer learning from simulation to real world via graph convolutional neural network (CNN) - by abstracting partially observable mobile network into graph, then distilling domain-variant irregular graph into domain-invariant tensor in locally Euclidean space as input to CNN -, domain randomization and multi-task learning. We use a novel self-play mechanism to encourage competition among DRL agents for best record on multiple tasks via simulated annealing, just like athletes compete for world record in decathlon. We also propose a decentralized multi-agent, competitive and cooperative DRL method to coordinate the actions of multi-cells to maximize global reward and minimize negative impact to neighbor cells. Using 6 field trials on commercial mobile networks, we demonstrate for the first time that a DRL agent can successfully transfer learning from simulation to complex real world problem with imperfect information, complex rules/dynamics, huge state/action space, and multi-agent interactions, without any training in the real world.