Rui Lu

Rui Lu, Siping Shi, Yanting Liu et al.

As artificial intelligence continues to evolve, it is increasingly capable of handling a wide range of video analytics tasks with merely one large model. One of the key foundation technologies is the Segment Anything Model (SAM), which allows the video analytics tasks to be determined on the fly according to the input prompts from the user. However, achieving real-time response in video analytics applications is crucial for user experiences due to the limited communication and computation resources on the edge, especially with SAM, where users may continuously interact by adding or adjusting prompts. In this paper, we propose SAMEdge, a novel edge-cloud computing architecture designed to support SAM computations for edge users. SAMEdge integrates new modules on the edge and the cloud to maximize analytics accuracy under visual prompts and image prompts input with latency constraints. It addresses resource challenges associated with prompt encoding and image encoding by offering a visual prompt transformation algorithm for visual prompts and efficient workload partitioning for image encoding. SAMEdge is implemented by extending the open-source SAM project from Meta AI. We demonstrate the practical application of SAMEdge through a case study on a Visual Tour Guide application. Our evaluation indicates that SAMEdge significantly enhances the accuracy of the video analytics application under distinct network bandwidths across various prompts.

Zhenyu Hou, Xin Lv, Rui Lu et al.

Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks. However, existing approaches mainly rely on imitation learning and struggle to achieve effective test-time scaling. While reinforcement learning (RL) holds promise for enabling self-exploration, recent attempts yield modest improvements in complex reasoning. In this paper, we present T1 to scale RL by encouraging exploration and understand inference scaling. We first initialize the LLM using synthesized chain-of-thought data that integrates trial-and-error and self-verification. To scale RL training, we promote increased sampling diversity through oversampling. We demonstrate that T1 with open LLMs as its base exhibits inference scaling behavior and achieves superior performance on challenging math reasoning benchmarks. More importantly, we present a simple strategy to examine inference scaling, where increased inference budgets directly lead to T1's better performance without any additional verification.

Rui Lu, Runzhe Wang, Kaifeng Lyu et al. · tsinghua

Score-based diffusion models have achieved incredible performance in generating realistic images, audio, and video data. While these models produce high-quality samples with impressive details, they often introduce unrealistic artifacts, such as distorted fingers or hallucinated texts with no meaning. This paper focuses on textual hallucinations, where diffusion models correctly generate individual symbols but assemble them in a nonsensical manner. Through experimental probing, we consistently observe that such phenomenon is attributed it to the network's local generation bias. Denoising networks tend to produce outputs that rely heavily on highly correlated local regions, particularly when different dimensions of the data distribution are nearly pairwise independent. This behavior leads to a generation process that decomposes the global distribution into separate, independent distributions for each symbol, ultimately failing to capture the global structure, including underlying grammar. Intriguingly, this bias persists across various denoising network architectures including MLP and transformers which have the structure to model global dependency. These findings also provide insights into understanding other types of hallucinations, extending beyond text, as a result of implicit biases in the denoising models. Additionally, we theoretically analyze the training dynamics for a specific case involving a two-layer MLP learning parity points on a hypercube, offering an explanation of its underlying mechanism.

Rui Lu, Bihai Zhang, Dan Wang

With the popularity of 3D volumetric video applications, such as Autonomous Driving, Virtual Reality, and Mixed Reality, current developers have turned to deep learning for compressing volumetric video frames, i.e., point clouds for video upstreaming. The latest deep learning-based solutions offer higher efficiency, lower distortion, and better hardware support compared to traditional ones like MPEG and JPEG. However, privacy threats arise, especially reconstruction attacks targeting to recover the original input point cloud from the intermediate results. In this paper, we design VVRec, to the best of our knowledge, which is the first targeting DL-based Volumetric Video Reconstruction attack scheme. VVRec demonstrates the ability to reconstruct high-quality point clouds from intercepted transmission intermediate results using four well-trained neural network modules we design. Leveraging the latest latent diffusion models with Gamma distribution and a refinement algorithm, VVRec excels in reconstruction quality, color recovery, and surpasses existing defenses. We evaluate VVRec using three volumetric video datasets. The results demonstrate that VVRec achieves 64.70dB reconstruction accuracy, with an impressive 46.39% reduction of distortion over baselines.