arXiv subjects

De Ma

Publications and source records attributed to De Ma.

17 recordsLinked to original sources

arXiv2026

DLAM: Distributional Latent Actions with Temporal Constraints

Vision-language-action (VLA) models remain constrained by scarce action-labeled robot data, whereas action-free videos offer abundant observations of physical change. Latent action models can extract such priors, but reconstruction-trained codes may predict future observations without the structure required for joint generation with robot actions. Existing structured methods add temporal constraints but retain deterministic transition points, so residual errors in locally inferred transitions may propagate and compound under recursive composition. We introduce DLAM, a distributional latent-action model that represents each transition as a diagonal Gaussian. Reconstruction conditioned on the reference frame grounds the mean in observed visual change, while normalized composition and reversal over equal-gap triplets constrain both the mean and dimension-wise variance. Variance composition uses a lightweight shared-correlation coefficient to account for dependence between adjacent transitions that share an intermediate frame, whereas reversal negates the mean and preserves the variance. For downstream policy learning, we freeze the encoder and train a flow-matching policy to jointly generate mean transition sequences and robot actions. On held-out transitions, DLAM learns more temporally consistent latent dynamics than existing latent-action baselines and achieves stronger direct and cumulative reconstruction on held-out videos. Under the same controlled $\pi_0$ transfer protocol, it also improves policy performance on MetaWorld MT50, LIBERO, and real-world manipulation tasks. Controlled ablations show that normalized mean constraints account for most of the reconstruction gain, while learned variance and correlation-aware composition provide complementary improvements in downstream control.

cs.RO↗

arXiv2026

ALAM: Algebraically Consistent Latent Action Model for Vision-Language-Action Models

Vision-language-action (VLA) models remain constrained by the scarcity of action-labeled robot data, whereas action-free videos provide abundant evidence of how the physical world changes. Latent action models offer a promising way to extract such priors from videos, but reconstruction-trained latent codes are not necessarily suitable for policy generation: they may predict future observations while lacking the structure needed to be reused or generated coherently with robot actions. We introduce ALAM (Algebraic Latent Action Model), an Algebraically Consistent Latent Action Model that turns temporal relations in action-free video into structural supervision. Given frame triplets, ALAM learns latent transitions that are grounded by reconstruction while being regularized by composition and reversal consistency, encouraging a locally additive transition space. For downstream VLA learning, we freeze the pretrained encoder and use its latent transition sequences as auxiliary generative targets, co-generated with robot actions under a joint flow-matching objective. This couples structured latent transitions with flow-based policy generation, allowing the policy to exploit ALAM's locally consistent transition geometry without requiring latent-to-action decoding. Representation probes show that ALAM reduces additivity and reversibility errors by 25-85 times over unstructured latent-action baselines and improves long-horizon cumulative reconstruction. When transferred to VLA policies, ALAM raises the average success rate from 47.9% to 85.0% on MetaWorld MT50 and from 94.1% to 98.1% on LIBERO, with consistent gains on real-world manipulation tasks. Ablations further confirm that the strongest improvements arise from the synergy between algebraically structured latent transitions and joint flow matching.

cs.RO↗

arXiv2026

One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy

Vision-language-action (VLA) models increasingly rely on auxiliary world modules to plan over long horizons, yet how such modules should be parameterized on top of a pretrained VLA remains an open design question. Existing world-model-augmented VLAs typically pass the per-frame visual stream into the world module at high visual bandwidth and treat its rollout as a side product of action prediction; under a constrained adaptation budget on a frozen backbone, this leaves both the per-frame representation and the latent action coupling under-examined. We introduce OneWM-VLA, which compresses each view into a single semantic token per frame through an Adaptive Attention Pooling, and produces the resulting latent stream and the action trajectory under a single flow-matching objective rather than connecting them through a separate decoder. Empirically, we find that per-frame visual bandwidth can be reduced to a single token without compromising long-horizon performance under our setup. Trained with 14.71M LoRA parameters on a $\pi_0$ (2B) backbone, OneWM-VLA improves the average success rate from 47.9% to 61.3% on MetaWorld~MT50, reaches 95.6% on LIBERO-Long (vs.85.2% for $\pi_0$), and reaches 60.0% on the long-horizon deformable task Fold Cloth on a real Piper arm (vs.20.0% for $\pi_0$).

cs.CV↗

arXiv2025

DarwinWafer: A Wafer-Scale Neuromorphic Chip

Neuromorphic computing promises brain-like efficiency, yet today's multi-chip systems scale over PCBs and incur orders-of-magnitude penalties in bandwidth, latency, and energy, undermining biological algorithms and system efficiency. We present DarwinWafer, a hyperscale system-on-wafer that replaces off-chip interconnects with wafer-scale, high-density integration of 64 Darwin3 chiplets on a 300 mm silicon interposer. A GALS NoC within each chiplet and an AER-based asynchronous wafer fabric with hierarchical time-step synchronization provide low-latency, coherent operation across the wafer. Each chiplet implements 2.35 M neurons and 0.1 B synapses, yielding 0.15 B neurons and 6.4 B synapses per wafer.At 333 MHz and 0.8 V, DarwinWafer consumes ~100 W and achieves 4.9 pJ/SOP, with 64 TSOPS peak throughput (0.64 TSOPS/W). Realization is enabled by a holistic chiplet-interposer co-design flow (including an in-house interposer-bump planner with early SI/PI and electro-thermal closure) and a warpage-tolerant assembly that fans out I/O via PCBlets and compliant pogo-pin connections, enabling robust, demountable wafer-to-board integration. Measurements confirm 10 mV supply droop and a uniform thermal profile (34-36 {\deg}C) under ~100 W. Application studies demonstrate whole-brain simulations: two zebrafish brains per chiplet with high connectivity fidelity (Spearman r = 0.896) and a mouse brain mapped across 32 chiplets (r = 0.645). To our knowledge, DarwinWafer represents a pioneering demonstration of wafer-scale neuromorphic computing, establishing a viable and scalable path toward large-scale, brain-like computation on silicon by replacing PCB-level interconnects with high-density, on-wafer integration.

cs.ET↗

arXiv2025

Bridging Quantized Artificial Neural Networks and Neuromorphic Hardware

Neuromorphic hardware aims to leverage distributed computing and event-driven circuit design to achieve an energy-efficient AI system. The name "neuromorphic" is derived from its spiking and local computing nature, which mimics the fundamental activity of an animal's nervous system. In neuromorphic hardware, neurons, i.e., computing cores use single-bit, event-driven data (called spikes) for inter-communication, which differs substantially from conventional hardware. To leverage the advantages of neuromorphic hardware and implement a computing model, the conventional approach is to build spiking neural networks (SNNs). SNNs replace the nonlinearity part of artificial neural networks (ANNs) in the realm of deep learning with spiking neurons, where the spiking neuron mimics the basic behavior of bio-neurons. However, there is still a performance gap between SNNs and their ANN counterparts. In this paper, we explore a new way to map computing models onto neuromorphic hardware. We propose a Spiking-Driven ANN (SDANN) framework that directly implements quantized ANN on hardware, eliminating the need for tuning the trainable parameters or any performance degradation. With the power of quantized ANN, our SDANN ensures a lower bound of implementation performance on neuromorphic hardware. To address the limitation of bit width support on hardware, we propose bias calibration and scaled integration methods. Experiments on various tasks demonstrate that our SDANN achieves exactly the same accuracy as the quantized ANN. Beyond toy examples and software implementation, we successfully deployed and validated our spiking models on real neuromorphic hardware, demonstrating the feasibility of the SDANN framework.

cs.NE↗

arXiv2025

Bidirectional Distillation: A Mixed-Play Framework for Multi-Agent Generalizable Behaviors

Population-population generalization is a challenging problem in multi-agent reinforcement learning (MARL), particularly when agents encounter unseen co-players. However, existing self-play-based methods are constrained by the limitation of inside-space generalization. In this study, we propose Bidirectional Distillation (BiDist), a novel mixed-play framework, to overcome this limitation in MARL. BiDist leverages knowledge distillation in two alternating directions: forward distillation, which emulates the historical policies' space and creates an implicit self-play, and reverse distillation, which systematically drives agents towards novel distributions outside the known policy space in a non-self-play manner. In addition, BiDist operates as a concise and efficient solution without the need for the complex and costly storage of past policies. We provide both theoretical analysis and empirical evidence to support BiDist's effectiveness. Our results highlight its remarkable generalization ability across a variety of cooperative, competitive, and social dilemma tasks, and reveal that BiDist significantly diversifies the policy distribution space. We also present comprehensive ablation studies to reinforce BiDist's effectiveness and key success factors. Source codes are available in the supplementary material.

cs.LG↗

arXiv2025

FpgaHub: Fpga-centric Hyper-heterogeneous Computing Platform for Big Data Analytics

Modern data analytics requires a huge amount of computing power and processes a massive amount of data. At the same time, the underlying computing platform is becoming much more heterogeneous on both hardware and software. Even though specialized hardware, e.g., FPGA- or GPU- or TPU-based systems, often achieves better performance than a CPU-only system due to the slowing of Moore's law, such systems are limited in what they can do. For example, GPU-only approaches suffer from severe IO limitations. To truly exploit the potential of hardware heterogeneity, we present FpgaHub, an FPGA-centric hyper-heterogeneous computing platform for big data analytics. The key idea of FpgaHub is to use reconfigurable computing to implement a versatile hub complementing other processors (CPUs, GPUs, DPUs, programmable switches, computational storage, etc.). Using an FPGA as the basis, we can take advantage of its highly reconfigurable nature and rich IO interfaces such as PCIe, networking, and on-board memory, to place it at the center of the architecture and use it as a data and control plane for data movement, scheduling, pre-processing, etc. FpgaHub enables architectural flexibility to allow exploring the rich design space of heterogeneous computing platforms.

cs.DC↗

arXiv2024

VLASCD: A Visual Language Action Model for Simultaneous Chatting and Decision Making

Recent large pretrained models such as LLMs (e.g., GPT series) and VLAs (e.g., OpenVLA) have achieved notable progress on multimodal tasks, yet they are built upon a multi-input single-output (MISO) paradigm. We show that this paradigm fundamentally limits performance in multi-input multi-output (MIMO) scenarios, where parallel task execution is required. In MISO architectures, tasks compete for a shared output channel, creating mutual exclusion effects that cause unbalanced optimization and degraded performance. To address this gap, we introduce MIMO-VLA (VLASCD), a unified training framework that enables concurrent multi-task outputs, exemplified by simultaneous dialogue generation and decision-making. Inspired by human cognition, MIMO-VLA eliminates interference between tasks and supports efficient parallel processing. Experiments on the CARLA autonomous driving platform demonstrate that MIMO-VLA substantially outperforms state-of-the-art MISO-based LLMs, reinforcement learning models, and VLAs in MIMO settings, establishing a new direction for multimodal and multitask learning.

cs.AI↗

arXiv2024

Spiking GS: Towards High-Accuracy and Low-Cost Surface Reconstruction via Spiking Neuron-based Gaussian Splatting

3D Gaussian Splatting is capable of reconstructing 3D scenes in minutes. Despite recent advances in improving surface reconstruction accuracy, the reconstructed results still exhibit bias and suffer from inefficiency in storage and training. This paper provides a different observation on the cause of the inefficiency and the reconstruction bias, which is attributed to the integration of the low-opacity parts (LOPs) of the generated Gaussians. We show that LOPs consist of Gaussians with overall low-opacity (LOGs) and the low-opacity tails (LOTs) of Gaussians. We propose Spiking GS to reduce such two types of LOPs by integrating spiking neurons into the Gaussian Splatting pipeline. Specifically, we introduce global and local full-precision integrate-and-fire spiking neurons to the opacity and representation function of flattened 3D Gaussians, respectively. Furthermore, we enhance the density control strategy with spiking neurons' thresholds and a new criterion on the scale of Gaussians. Our method can represent more accurate reconstructed surfaces at a lower cost. The supplementary material and code are available at https://github.com/zju-bmi-lab/SpikingGS.

cs.CV↗

arXiv2024

An Asynchronous Multi-core Accelerator for SNN inference

Spiking Neural Networks (SNNs) are extensively utilized in brain-inspired computing and neuroscience research. To enhance the speed and energy efficiency of SNNs, several many-core accelerators have been developed. However, maintaining the accuracy of SNNs often necessitates frequent explicit synchronization among all cores, which presents a challenge to overall efficiency. In this paper, we propose an asynchronous architecture for Spiking Neural Networks (SNNs) that eliminates the need for inter-core synchronization, thus enhancing speed and energy efficiency. This approach leverages the pre-determined dependencies of neuromorphic cores established during compilation. Each core is equipped with a scheduler that monitors the status of its dependencies, allowing it to safely advance to the next timestep without waiting for other cores. This eliminates the necessity for global synchronization and minimizes core waiting time despite inherent workload imbalances. Comprehensive evaluations using five different SNN workloads show that our architecture achieves a 1.86x speedup and a 1.55x increase in energy efficiency compared to state-of-the-art synchronization architectures.

cs.NE↗

arXiv2024

EAS-SNN: End-to-End Adaptive Sampling and Representation for Event-based Detection with Recurrent Spiking Neural Networks

Event cameras, with their high dynamic range and temporal resolution, are ideally suited for object detection, especially under scenarios with motion blur and challenging lighting conditions. However, while most existing approaches prioritize optimizing spatiotemporal representations with advanced detection backbones and early aggregation functions, the crucial issue of adaptive event sampling remains largely unaddressed. Spiking Neural Networks (SNNs), which operate on an event-driven paradigm through sparse spike communication, emerge as a natural fit for addressing this challenge. In this study, we discover that the neural dynamics of spiking neurons align closely with the behavior of an ideal temporal event sampler. Motivated by this insight, we propose a novel adaptive sampling module that leverages recurrent convolutional SNNs enhanced with temporal memory, facilitating a fully end-to-end learnable framework for event-based detection. Additionally, we introduce Residual Potential Dropout (RPD) and Spike-Aware Training (SAT) to regulate potential distribution and address performance degradation encountered in spike-based sampling modules. Empirical evaluation on neuromorphic detection datasets demonstrates that our approach outperforms existing state-of-the-art spike-based methods with significantly fewer parameters and time steps. For instance, our method yields a 4.4\% mAP improvement on the Gen1 dataset, while requiring 38\% fewer parameters and only three time steps. Moreover, the applicability and effectiveness of our adaptive sampling methodology extend beyond SNNs, as demonstrated through further validation on conventional non-spiking models. Code is available at https://github.com/Windere/EAS-SNN.

cs.CV↗

arXiv2024

Learning to Manipulate Artistic Images

Recent advancement in computer vision has significantly lowered the barriers to artistic creation. Exemplar-based image translation methods have attracted much attention due to flexibility and controllability. However, these methods hold assumptions regarding semantics or require semantic information as the input, while accurate semantics is not easy to obtain in artistic images. Besides, these methods suffer from cross-domain artifacts due to training data prior and generate imprecise structure due to feature compression in the spatial domain. In this paper, we propose an arbitrary Style Image Manipulation Network (SIM-Net), which leverages semantic-free information as guidance and a region transportation strategy in a self-supervised manner for image generation. Our method balances computational efficiency and high resolution to a certain extent. Moreover, our method facilitates zero-shot style image manipulation. Both qualitative and quantitative experiments demonstrate the superiority of our method over state-of-the-art methods.Code is available at https://github.com/SnailForce/SIM-Net.

cs.CV↗

arXiv2023

Darwin3: A large-scale neuromorphic chip with a Novel ISA and On-Chip Learning

Spiking Neural Networks (SNNs) are gaining increasing attention for their biological plausibility and potential for improved computational efficiency. To match the high spatial-temporal dynamics in SNNs, neuromorphic chips are highly desired to execute SNNs in hardware-based neuron and synapse circuits directly. This paper presents a large-scale neuromorphic chip named Darwin3 with a novel instruction set architecture(ISA), which comprises 10 primary instructions and a few extended instructions. It supports flexible neuron model programming and local learning rule designs. The Darwin3 chip architecture is designed in a mesh of computing nodes with an innovative routing algorithm. We used a compression mechanism to represent synaptic connections, significantly reducing memory usage. The Darwin3 chip supports up to 2.35 million neurons, making it the largest of its kind in neuron scale. The experimental results showed that code density was improved up to 28.3x in Darwin3, and neuron core fan-in and fan-out were improved up to 4096x and 3072x by connection compression compared to the physical memory depth. Our Darwin3 chip also provided memory saving between 6.8X and 200.8X when mapping convolutional spiking neural networks (CSNN) onto the chip, demonstrating state-of-the-art performance in accuracy and latency compared to other neuromorphic chips.

cs.NE↗

arXiv2022

Multi-Level Firing with Spiking DS-ResNet: Enabling Better and Deeper Directly-Trained Spiking Neural Networks

Spiking neural networks (SNNs) are bio-inspired neural networks with asynchronous discrete and sparse characteristics, which have increasingly manifested their superiority in low energy consumption. Recent research is devoted to utilizing spatio-temporal information to directly train SNNs by backpropagation. However, the binary and non-differentiable properties of spike activities force directly trained SNNs to suffer from serious gradient vanishing and network degradation, which greatly limits the performance of directly trained SNNs and prevents them from going deeper. In this paper, we propose a multi-level firing (MLF) method based on the existing spatio-temporal back propagation (STBP) method, and spiking dormant-suppressed residual network (spiking DS-ResNet). MLF enables more efficient gradient propagation and the incremental expression ability of the neurons. Spiking DS-ResNet can efficiently perform identity mapping of discrete spikes, as well as provide a more suitable connection for gradient propagation in deep SNNs. With the proposed method, our model achieves superior performances on a non-neuromorphic dataset and two neuromorphic datasets with much fewer trainable parameters and demonstrates the great ability to combat the gradient vanishing and degradation problem in deep SNNs.

cs.NE↗

arXiv2018

Efficient Spiking Neural Networks with Logarithmic Temporal Coding

A Spiking Neural Network (SNN) can be trained indirectly by first training an Artificial Neural Network (ANN) with the conventional backpropagation algorithm, then converting it into an SNN. The conventional rate-coding method for SNNs uses the number of spikes to encode magnitude of an activation value, and may be computationally inefficient due to the large number of spikes. Temporal-coding is typically more efficient by leveraging the timing of spikes to encode information. In this paper, we present Logarithmic Temporal Coding (LTC), where the number of spikes used to encode an activation value grows logarithmically with the activation value; and the accompanying Exponentiate-and-Fire (EF) spiking neuron model, which only involves efficient bit-shift and addition operations. Moreover, we improve the training process of ANN to compensate for approximation errors due to LTC. Experimental results indicate that the resulting SNN achieves competitive performance at significantly lower computational cost than related work.

cs.NE↗

arXiv2018

Dating Ancient Paintings of Mogao Grottoes Using Deeply Learnt Visual Codes

Cultural heritage is the asset of all the peoples of the world. The preservation and inheritance of cultural heritage is conducive to the progress of human civilization. In northwestern China, there is a world heritage site -- Mogao Grottoes -- that has a plenty of mural paintings showing the historical cultures of ancient China. To study these historical cultures, one critical procedure is to date the mural paintings, i.e., determining the era when they were created. Until now, most mural paintings at Mogao Grottoes have been dated by directly referring to the mural texts or historical documents. However, some are still left with creation-era undetermined due to the lack of reference materials. Considering that the drawing style of mural paintings was changing along the history and the drawing style can be learned and quantified through painting data, we formulate the problem of mural-painting dating into a problem of drawing-style classification. In fact, drawing styles can be expressed not only in color or curvature, but also in some unknown forms -- the forms that have not been observed. To this end, besides sophisticated color and shape descriptors, a deep convolution neural network is designed to encode the implicit drawing styles. 3860 mural paintings collected from 194 different grottoes with determined creation-era labels are used to train the classification model and build the dating method. In experiments, the proposed dating method is applied to seven mural paintings which were previously dated with controversies, and the exciting new dating results are approved by the Dunhuang expert.

cs.CV↗

arXiv2018

Scalable NoC-based Neuromorphic Hardware Learning and Inference

Bio-inspired neuromorphic hardware is a research direction to approach brain's computational power and energy efficiency. Spiking neural networks (SNN) encode information as sparsely distributed spike trains and employ spike-timing-dependent plasticity (STDP) mechanism for learning. Existing hardware implementations of SNN are limited in scale or do not have in-hardware learning capability. In this work, we propose a low-cost scalable Network-on-Chip (NoC) based SNN hardware architecture with fully distributed in-hardware STDP learning capability. All hardware neurons work in parallel and communicate through the NoC. This enables chip-level interconnection, scalability and reconfigurability necessary for deploying different applications. The hardware is applied to learn MNIST digits as an evaluation of its learning capability. We explore the design space to study the trade-offs between speed, area and energy. How to use this procedure to find optimal architecture configuration is also discussed.

cs.ET↗