Bin Zhu

Sungwon Han, Sungwon Park, Fangzhao Wu et al.

Federated learning is used to train a shared model in a decentralized way without clients sharing private data with each other. Federated learning systems are susceptible to poisoning attacks when malicious clients send false updates to the central server. Existing defense strategies are ineffective under non-IID data settings. This paper proposes a new defense strategy, FedCPA (Federated learning with Critical Parameter Analysis). Our attack-tolerant aggregation method is based on the observation that benign local models have similar sets of top-k and bottom-k critical parameters, whereas poisoned local models do not. Experiments with different attack scenarios on multiple datasets demonstrate that our model outperforms existing defense strategies in defending against poisoning attacks.

Chengming Feng, Jing Hu, Xin Wang et al.

Controlling the degree of stylization in the Neural Style Transfer (NST) is a little tricky since it usually needs hand-engineering on hyper-parameters. In this paper, we propose the first deep Reinforcement Learning (RL) based architecture that splits one-step style transfer into a step-wise process for the NST task. Our RL-based method tends to preserve more details and structures of the content image in early steps, and synthesize more style patterns in later steps. It is a user-easily-controlled style-transfer method. Additionally, as our RL-based model performs the stylization progressively, it is lightweight and has lower computational complexity than existing one-step Deep Learning (DL) based models. Experimental results demonstrate the effectiveness and robustness of our method.

Bin Lin, Yunyang Ge, Xinhua Cheng et al.

We introduce Open-Sora Plan, an open-source project that aims to contribute a large generation model for generating desired high-resolution videos with long durations based on various user inputs. Our project comprises multiple components for the entire video generation process, including a Wavelet-Flow Variational Autoencoder, a Joint Image-Video Skiparse Denoiser, and various condition controllers. Moreover, many assistant strategies for efficient training and inference are designed, and a multi-dimensional data curation pipeline is proposed for obtaining desired high-quality data. Benefiting from efficient thoughts, our Open-Sora Plan achieves impressive video generation results in both qualitative and quantitative evaluations. We hope our careful design and practical experience can inspire the video generation research community. All our codes and model weights are publicly available at \url{https://github.com/PKU-YuanGroup/Open-Sora-Plan}.

Jingwei Yi, Yueqi Xie, Bin Zhu et al.

The integration of large language models with external content has enabled applications such as Microsoft Copilot but also introduced vulnerabilities to indirect prompt injection attacks. In these attacks, malicious instructions embedded within external content can manipulate LLM outputs, causing deviations from user expectations. To address this critical yet under-explored issue, we introduce the first benchmark for indirect prompt injection attacks, named BIPIA, to assess the risk of such vulnerabilities. Using BIPIA, we evaluate existing LLMs and find them universally vulnerable. Our analysis identifies two key factors contributing to their success: LLMs' inability to distinguish between informational context and actionable instructions, and their lack of awareness in avoiding the execution of instructions within external content. Based on these findings, we propose two novel defense mechanisms-boundary awareness and explicit reminder-to address these vulnerabilities in both black-box and white-box settings. Extensive experiments demonstrate that our black-box defense provides substantial mitigation, while our white-box defense reduces the attack success rate to near-zero levels, all while preserving the output quality of LLMs. We hope this work inspires further research into securing LLM applications and fostering their safe and reliable use.

Toby Perrett, Ahmad Darkhalil, Saptarshi Sinha et al.

We present a validation dataset of newly-collected kitchen-based egocentric videos, manually annotated with highly detailed and interconnected ground-truth labels covering: recipe steps, fine-grained actions, ingredients with nutritional values, moving objects, and audio annotations. Importantly, all annotations are grounded in 3D through digital twinning of the scene, fixtures, object locations, and primed with gaze. Footage is collected from unscripted recordings in diverse home environments, making HDEPIC the first dataset collected in-the-wild but with detailed annotations matching those in controlled lab environments. We show the potential of our highly-detailed annotations through a challenging VQA benchmark of 26K questions assessing the capability to recognise recipes, ingredients, nutrition, fine-grained actions, 3D perception, object motion, and gaze direction. The powerful long-context Gemini Pro only achieves 38.5% on this benchmark, showcasing its difficulty and highlighting shortcomings in current VLMs. We additionally assess action recognition, sound recognition, and long-term video-object segmentation on HD-EPIC. HD-EPIC is 41 hours of video in 9 kitchens with digital twins of 413 kitchen fixtures, capturing 69 recipes, 59K fine-grained actions, 51K audio events, 20K object movements and 37K object masks lifted to 3D. On average, we have 263 annotations per minute of our unscripted videos.

Jielin Song, Siyu Liu, Bin Zhu et al.

As large language models (LLMs) continue to advance, instruction tuning has become critical for improving their ability to generate accurate and contextually appropriate responses. Although numerous instruction-tuning datasets have been developed to enhance LLM performance, selecting high-quality instruction data from large source datasets typically demands significant human effort. In this work, we introduce $\textbf{IterSelectTune}$, an efficient, cost-effective iterative training policy for selecting high-quality instruction data with no human involvement and limited reliance on GPT-4. By fine-tuning on approximately 20\% of the source data, our method consistently outperforms models fine-tuned on the full dataset across multiple benchmarks and public test datasets. These results highlight the effectiveness of our approach in enhancing LLM performance while reducing the computational resources required for instruction tuning.

Wenjun Peng, Jingwei Yi, Fangzhao Wu et al.

Large language models (LLMs) have demonstrated powerful capabilities in both text understanding and generation. Companies have begun to offer Embedding as a Service (EaaS) based on these LLMs, which can benefit various natural language processing (NLP) tasks for customers. However, previous studies have shown that EaaS is vulnerable to model extraction attacks, which can cause significant losses for the owners of LLMs, as training these models is extremely expensive. To protect the copyright of LLMs for EaaS, we propose an Embedding Watermark method called EmbMarker that implants backdoors on embeddings. Our method selects a group of moderate-frequency words from a general text corpus to form a trigger set, then selects a target embedding as the watermark, and inserts it into the embeddings of texts containing trigger words as the backdoor. The weight of insertion is proportional to the number of trigger words included in the text. This allows the watermark backdoor to be effectively transferred to EaaS-stealer's model for copyright verification while minimizing the adverse impact on the original embeddings' utility. Our extensive experiments on various datasets show that our method can effectively protect the copyright of EaaS models without compromising service quality.

Bin Zhu, Zhaoquan Gu, Le Wang et al.

Recent work shows that deep neural networks are vulnerable to adversarial examples. Much work studies adversarial example generation, while very little work focuses on more critical adversarial defense. Existing adversarial detection methods usually make assumptions about the adversarial example and attack method (e.g., the word frequency of the adversarial example, the perturbation level of the attack method). However, this limits the applicability of the detection method. To this end, we propose TREATED, a universal adversarial detection method that can defend against attacks of various perturbation levels without making any assumptions. TREATED identifies adversarial examples through a set of well-designed reference models. Extensive experiments on three competitive neural networks and two widely used datasets show that our method achieves better detection performance than baselines. We finally conduct ablation studies to verify the effectiveness of our method.

Quanyu Liao, Xin Wang, Bin Kong et al.

Deep neural networks have been demonstrated to be vulnerable to adversarial attacks: subtle perturbation can completely change prediction result. The vulnerability has led to a surge of research in this direction, including adversarial attacks on object detection networks. However, previous studies are dedicated to attacking anchor-based object detectors. In this paper, we present the first adversarial attack on anchor-free object detectors. It conducts category-wise, instead of previously instance-wise, attacks on object detectors, and leverages high-level semantic information to efficiently generate transferable adversarial examples, which can also be transferred to attack other object detectors, even anchor-based detectors such as Faster R-CNN. Experimental results on two benchmark datasets demonstrate that our proposed method achieves state-of-the-art performance and transferability.

Quanyu Liao, Yuezun Li, Xin Wang et al.

Fooling people with highly realistic fake images generated with Deepfake or GANs brings a great social disturbance to our society. Many methods have been proposed to detect fake images, but they are vulnerable to adversarial perturbations -- intentionally designed noises that can lead to the wrong prediction. Existing methods of attacking fake image detectors usually generate adversarial perturbations to perturb almost the entire image. This is redundant and increases the perceptibility of perturbations. In this paper, we propose a novel method to disrupt the fake image detection by determining key pixels to a fake image detector and attacking only the key pixels, which results in the $L_0$ and the $L_2$ norms of adversarial perturbations much less than those of existing works. Experiments on two public datasets with three fake image detectors indicate that our proposed method achieves state-of-the-art performance in both white-box and black-box attacks.

Yiming Li, Shan Liu, Yu Chen et al.

As the successor of H.265/HEVC, the new versatile video coding standard (H.266/VVC) can provide up to 50% bitrate saving with the same subjective quality, at the cost of increased decoding complexity. To accelerate the application of the new coding standard, a real-time H.266/VVC software decoder that can support various platforms is implemented, where SIMD technologies, parallelism optimization, and the acceleration strategies based on the characteristics of each coding tool are applied. As the mobile devices have become an essential carrier for video services nowadays, the mentioned optimization efforts are not only implemented for the x86 platform, but more importantly utilized to highly optimize the decoding performance on the ARM platform in this work. The experimental results show that when running on the Apple A14 SoC (iPhone 12pro), the average single-thread decoding speed of the present implementation can achieve 53fps (RA and LB) for full HD (1080p) bitstreams generated by VTM-11.0 reference software using 8bit Common Test Conditions (CTC). When multi-threading is enabled, an average of 32 fps (RA) can be achieved when decoding the 4K bitstreams.