Hayden Kwok‐Hay So

Jiajun Zhou, Jiajun Wu, Yizhao Gao et al.

To accelerate the inference of deep neural networks (DNNs), quantization with low-bitwidth numbers is actively researched. A prominent challenge is to quantize the DNN models into low-bitwidth numbers without significant accuracy degradation, especially at very low bitwidths (< 8 bits). This work targets an adaptive data representation with variable-length encoding called DyBit. DyBit can dynamically adjust the precision and range of separate bit-field to be adapted to the DNN weights/activations distribution. We also propose a hardware-aware quantization framework with a mixed-precision accelerator to trade-off the inference accuracy and speedup. Experimental results demonstrate that the inference accuracy via DyBit is 1.997% higher than the state-of-the-art at 4-bit quantization, and the proposed framework can achieve up to 8.1x speedup compared with the original model.

Song Wang, Zhu Wang, Can Li et al.

In comparison to conventional RGB cameras, the superior temporal resolution of event cameras allows them to capture rich information between frames, making them prime candidates for object tracking. Yet in practice, despite their theoretical advantages, the body of work on event-based multi-object tracking (MOT) remains in its infancy, especially in real-world settings where events from complex background and camera motion can easily obscure the true target motion. In this work, an event-based multi-object tracker, called SpikeMOT, is presented to address these challenges. SpikeMOT leverages spiking neural networks to extract sparse spatiotemporal features from event streams associated with objects. The resulting spike train representations are used to track the object movement at high frequency, while a simultaneous object detector provides updated spatial information of these objects at an equivalent frame rate. To evaluate the effectiveness of SpikeMOT, we introduce DSEC-MOT, the first large-scale event-based MOT benchmark incorporating fine-grained annotations for objects experiencing severe occlusions, frequent trajectory intersections, and long-term re-identification in real-world contexts. Extensive experiments employing DSEC-MOT and another event-based dataset, named FE240hz, demonstrate SpikeMOT's capability to achieve high tracking accuracy amidst challenging real-world scenarios, advancing the state-of-the-art in event-based multi-object tracking.

Yuan Chang, Ziyue Li, Hengyuan Zhang et al.

While Large Language Models (LLMs) have shown significant potential in assisting peer review, current methods often struggle to generate thorough and insightful reviews while maintaining efficiency. In this paper, we propose TreeReview, a novel framework that models paper review as a hierarchical and bidirectional question-answering process. TreeReview first constructs a tree of review questions by recursively decomposing high-level questions into fine-grained sub-questions and then resolves the question tree by iteratively aggregating answers from leaf to root to get the final review. Crucially, we incorporate a dynamic question expansion mechanism to enable deeper probing by generating follow-up questions when needed. We construct a benchmark derived from ICLR and NeurIPS venues to evaluate our method on full review generation and actionable feedback comments generation tasks. Experimental results of both LLM-based and human evaluation show that TreeReview outperforms strong baselines in providing comprehensive, in-depth, and expert-aligned review feedback, while reducing LLM token usage by up to 80% compared to computationally intensive approaches. Our code and benchmark dataset are available at https://github.com/YuanChang98/tree-review.

Baoheng Zhang, Yizhao Gao, Jingyuan Li et al.

Eye-tracking technology is integral to numerous consumer electronics applications, particularly in the realm of virtual and augmented reality (VR/AR). These applications demand solutions that excel in three crucial aspects: low-latency, low-power consumption, and precision. Yet, achieving optimal performance across all these fronts presents a formidable challenge, necessitating a balance between sophisticated algorithms and efficient backend hardware implementations. In this study, we tackle this challenge through a synergistic software/hardware co-design of the system with an event camera. Leveraging the inherent sparsity of event-based input data, we integrate a novel sparse FPGA dataflow accelerator customized for submanifold sparse convolution neural networks (SCNN). The SCNN implemented on the accelerator can efficiently extract the embedding feature vector from each representation of event slices by only processing the non-zero activations. Subsequently, these vectors undergo further processing by a gated recurrent unit (GRU) and a fully connected layer on the host CPU to generate the eye centers. Deployment and evaluation of our system reveal outstanding performance metrics. On the Event-based Eye-Tracking-AIS2024 dataset, our system achieves 81% p5 accuracy, 99.5% p10 accuracy, and 3.71 Mean Euclidean Distance with 0.7 ms latency while only consuming 2.29 mJ per inference. Notably, our solution opens up opportunities for future eye-tracking systems. Code is available at https://github.com/CASR-HKU/ESDA/tree/eye_tracking.

Yizhao Gao, Shuming Guo, Shijie Cao et al. · tsinghua

We introduce SeerAttention-R, a sparse attention framework specifically tailored for the long decoding of reasoning models. Extended from SeerAttention, SeerAttention-R retains the design of learning attention sparsity through a self-distilled gating mechanism, while removing query pooling to accommodate auto-regressive decoding. With a lightweight plug-in gating, SeerAttention-R is flexible and can be easily integrated into existing pretrained model without modifying the original parameters. We demonstrate that SeerAttention-R, trained on just 0.4B tokens, maintains near-lossless reasoning accuracy with 4K token budget in AIME benchmark under large sparse attention block sizes (64/128). Using TileLang, we develop a highly optimized sparse decoding kernel that achieves near-theoretical speedups of up to 9x over FlashAttention-3 on H100 GPU at 90% sparsity. Code is available at: https://github.com/microsoft/SeerAttention.

Hengyuan Zhang, Xinrong Chen, Yingmin Qiu et al.

Parameter-efficient fine-tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA), offer an efficient way to adapt large language models with reduced computational costs. However, their performance is limited by the small number of trainable parameters. Recent work combines LoRA with the Mixture-of-Experts (MoE), i.e., LoRA-MoE, to enhance capacity, but two limitations remain in hindering the full exploitation of its potential: 1) the influence of downstream tasks when assigning expert numbers, and 2) the uniform rank assignment across all LoRA experts, which restricts representational diversity. To mitigate these gaps, we propose GuiLoMo, a fine-grained layer-wise expert numbers and ranks allocation strategy with GuidedSelection Vectors (GSVs). GSVs are learned via a prior bilevel optimization process to capture both model- and task-specific needs, and are then used to allocate optimal expert numbers and ranks. Experiments on three backbone models across diverse benchmarks show that GuiLoMo consistently achieves superior or comparable performance to all baselines. Further analysis offers key insights into how expert numbers and ranks vary across layers and tasks, highlighting the benefits of adaptive expert configuration. Our code is available at https://github.com/Liar406/Gui-LoMo.git.

Maolin Wang, Seyedramin Rasoulinezhad, Philip H. W. Leong et al.

While integer arithmetic has been widely adopted for improved performance in deep quantized neural network inference, training remains a task primarily executed using floating point arithmetic. This is because both high dynamic range and numerical accuracy are central to the success of most modern training algorithms. However, due to its potential for computational, storage and energy advantages in hardware accelerators, neural network training methods that can be implemented with low precision integer-only arithmetic remains an active research challenge. In this paper, we present NITI, an efficient deep neural network training framework that stores all parameters and intermediate values as integers, and computes exclusively with integer arithmetic. A pseudo stochastic rounding scheme that eliminates the need for external random number generation is proposed to facilitate conversion from wider intermediate results to low precision storage. Furthermore, a cross-entropy loss backpropagation scheme computed with integer-only arithmetic is proposed. A proof-of-concept open-source software implementation of NITI that utilizes native 8-bit integer operations in modern GPUs to achieve end-to-end training is presented. When compared with an equivalent training setup implemented with floating point storage and arithmetic, NITI achieves negligible accuracy degradation on the MNIST and CIFAR10 datasets using 8-bit integer storage and computation. On ImageNet, 16-bit integers are needed for weight accumulation with an 8-bit datapath. This achieves training results comparable to all-floating-point implementations.

Zuowen Wang, Chang Gao, Zongwei Wu et al.

This survey reviews the AIS 2024 Event-Based Eye Tracking (EET) Challenge. The task of the challenge focuses on processing eye movement recorded with event cameras and predicting the pupil center of the eye. The challenge emphasizes efficient eye tracking with event cameras to achieve good task accuracy and efficiency trade-off. During the challenge period, 38 participants registered for the Kaggle competition, and 8 teams submitted a challenge factsheet. The novel and diverse methods from the submitted factsheets are reviewed and analyzed in this survey to advance future event-based eye tracking research.

Shaocong Wang, Yizhao Gao, Yi Li et al.

Visual sensors, including 3D LiDAR, neuromorphic DVS sensors, and conventional frame cameras, are increasingly integrated into edge-side intelligent machines. Realizing intensive multi-sensory data analysis directly on edge intelligent machines is crucial for numerous emerging edge applications, such as augmented and virtual reality and unmanned aerial vehicles, which necessitates unified data representation, unprecedented hardware energy efficiency and rapid model training. However, multi-sensory data are intrinsically heterogeneous, causing significant complexity in the system development for edge-side intelligent machines. In addition, the performance of conventional digital hardware is limited by the physically separated processing and memory units, known as the von Neumann bottleneck, and the physical limit of transistor scaling, which contributes to the slowdown of Moore's law. These limitations are further intensified by the tedious training of models with ever-increasing sizes. We propose a novel hardware-software co-design, random resistive memory-based deep extreme point learning machine (DEPLM), that offers efficient unified point set analysis. We show the system's versatility across various data modalities and two different learning tasks. Compared to a conventional digital hardware-based system, our co-design system achieves huge energy efficiency improvements and training cost reduction when compared to conventional systems. Our random resistive memory-based deep extreme point learning machine may pave the way for energy-efficient and training-friendly edge AI across various data modalities and tasks.

Jie Ran, Rui Lin, Hayden K. H. So et al.

Elasticities in depth, width, kernel size and resolution have been explored in compressing deep neural networks (DNNs). Recognizing that the kernels in a convolutional neural network (CNN) are 4-way tensors, we further exploit a new elasticity dimension along the input-output channels. Specifically, a novel nuclear-norm rank minimization factorization (NRMF) approach is proposed to dynamically and globally search for the reduced tensor ranks during training. Correlation between tensor ranks across multiple layers is revealed, and a graceful tradeoff between model size and accuracy is obtained. Experiments then show the superiority of NRMF over the previous non-elastic variational Bayesian matrix factorization (VBMF) scheme.

Zhen Dong, Yizhao Gao, Qijing Huang et al.

Automatic algorithm-hardware co-design for DNN has shown great success in improving the performance of DNNs on FPGAs. However, this process remains challenging due to the intractable search space of neural network architectures and hardware accelerator implementation. Differing from existing hardware-aware neural architecture search (NAS) algorithms that rely solely on the expensive learning-based approaches, our work incorporates integer programming into the search algorithm to prune the design space. Given a set of hardware resource constraints, our integer programming formulation directly outputs the optimal accelerator configuration for mapping a DNN subgraph that minimizes latency. We use an accuracy predictor for different DNN subgraphs with different quantization schemes and generate accuracy-latency pareto frontiers. With low computational cost, our algorithm can generate quantized networks that achieve state-of-the-art accuracy and hardware performance on Xilinx Zynq (ZU3EG) FPGA for image classification on ImageNet dataset. The solution searched by our algorithm achieves 72.5% top-1 accuracy on ImageNet at framerate 50, which is 60% faster than MnasNet and 135% faster than FBNet with comparable accuracy.

Sung-En Chang, Yanyu Li, Mengshu Sun et al.

Deep Neural Networks (DNNs) have achieved extraordinary performance in various application domains. To support diverse DNN models, efficient implementations of DNN inference on edge-computing platforms, e.g., ASICs, FPGAs, and embedded systems, are extensively investigated. Due to the huge model size and computation amount, model compression is a critical step to deploy DNN models on edge devices. This paper focuses on weight quantization, a hardware-friendly model compression approach that is complementary to weight pruning. Unlike existing methods that use the same quantization scheme for all weights, we propose the first solution that applies different quantization schemes for different rows of the weight matrix. It is motivated by (1) the distribution of the weights in the different rows are not the same; and (2) the potential of achieving better utilization of heterogeneous FPGA hardware resources. To achieve that, we first propose a hardware-friendly quantization scheme named sum-of-power-of-2 (SP2) suitable for Gaussian-like weight distribution, in which the multiplication arithmetic can be replaced with logic shifter and adder, thereby enabling highly efficient implementations with the FPGA LUT resources. In contrast, the existing fixed-point quantization is suitable for Uniform-like weight distribution and can be implemented efficiently by DSP. Then to fully explore the resources, we propose an FPGA-centric mixed scheme quantization (MSQ) with an ensemble of the proposed SP2 and the fixed-point schemes. Combining the two schemes can maintain, or even increase accuracy due to better matching with weight distributions.

Junjie Liu, Zhe Xu, Runbin Shi et al.

We present a novel network pruning algorithm called Dynamic Sparse Training that can jointly find the optimal network parameters and sparse network structure in a unified optimization process with trainable pruning thresholds. These thresholds can have fine-grained layer-wise adjustments dynamically via backpropagation. We demonstrate that our dynamic sparse training algorithm can easily train very sparse neural network models with little performance loss using the same number of training epochs as dense models. Dynamic Sparse Training achieves the state of the art performance compared with other sparse training algorithms on various network architectures. Additionally, we have several surprising observations that provide strong evidence for the effectiveness and efficiency of our algorithm. These observations reveal the underlying problems of traditional three-stage pruning algorithms and present the potential guidance provided by our algorithm to the design of more compact network architectures.

Nan Meng, Hayden K. -H. So, Xing Sun et al.

We consider the problem of high-dimensional light field reconstruction and develop a learning-based framework for spatial and angular super-resolution. Many current approaches either require disparity clues or restore the spatial and angular details separately. Such methods have difficulties with non-Lambertian surfaces or occlusions. In contrast, we formulate light field super-resolution (LFSR) as tensor restoration and develop a learning framework based on a two-stage restoration with 4-dimensional (4D) convolution. This allows our model to learn the features capturing the geometry information encoded in multiple adjacent views. Such geometric features vary near the occlusion regions and indicate the foreground object border. To train a feasible network, we propose a novel normalization operation based on a group of views in the feature maps, design a stage-wise loss function, and develop the multi-range training strategy to further improve the performance. Evaluations are conducted on a number of light field datasets including real-world scenes, synthetic data, and microscope light fields. The proposed method achieves superior performance and less execution time comparing with other state-of-the-art schemes.

Xing Sun, Nelson H. C. Yung, Edmund Y. Lam et al.

This technical report proves components consistency for the Doubly Stochastic Dirichlet Process with exponential convergence of posterior probability. We also present the fundamental properties for DSDP as well as inference algorithms. Simulation toy experiment and real-world experiment results for single and multi-cluster also support the consistency proof. This report is also a support document for the paper "Computationally Efficient Hyperspectral Data Learning Based on the Doubly Stochastic Dirichlet Process".