Zhiyong Feng

Zhigang Cen, Ningyan Guo, Wenjing Xu et al.

Video semantic segmentation(VSS) has been widely employed in lots of fields, such as simultaneous localization and mapping, autonomous driving and surveillance. Its core challenge is how to leverage temporal information to achieve better segmentation. Previous efforts have primarily focused on pixel-level static-dynamic contexts matching, utilizing techniques such as optical flow and attention mechanisms. Instead, this paper rethinks static-dynamic contexts at the class level and proposes a novel static-dynamic class-level perceptual consistency (SD-CPC) framework. In this framework, we propose multivariate class prototype with contrastive learning and a static-dynamic semantic alignment module. The former provides class-level constraints for the model, obtaining personalized inter-class features and diversified intra-class features. The latter first establishes intra-frame spatial multi-scale and multi-level correlations to achieve static semantic alignment. Then, based on cross-frame static perceptual differences, it performs two-stage cross-frame selective aggregation to achieve dynamic semantic alignment. Meanwhile, we propose a window-based attention map calculation method that leverages the sparsity of attention points during cross-frame aggregation to reduce computation cost. Extensive experiments on VSPW and Cityscapes datasets show that the proposed approach outperforms state-of-the-art methods. Our implementation will be open-sourced on GitHub.

Jingyuan Yang, Rongjun Li, Weixuan Wang et al.

Large Language Models (LLMs) often generate inconsistent responses when prompted with semantically equivalent paraphrased inputs. Recently, activation steering, a technique that modulates LLMs' behaviours by adjusting their latent representations during inference time, has been explored to improve the semantic consistency of LLMs. However, these methods typically operate at the model component level, such as layer hidden states or attention head outputs. They face a challenge due to the ``polysemanticity issue'', where the model components of LLMs typically encode multiple entangled features, making precise steering difficult. To address this challenge, we drill down to feature-level representations and propose LF-Steering, a novel activation steering approach to precisely identify latent feature representations responsible for semantic inconsistency. More specifically, our method maps the hidden states of the relevant transformer layer into a sparsely activated, high-dimensional feature space based on a sparse autoencoder (SAE), ensuring model steering based on decoupled feature representations with minimal interference. Comprehensive experiments on NLU and NLG datasets demonstrate the effectiveness of our method in enhancing semantic consistency, resulting in significant performance gains for various NLU and NLG tasks.

Minghui Zou, Ronghui Guo, Sai Zhang et al.

As the size and context length of Large Language Models (LLMs) grow, weight-activation quantization has emerged as a crucial technique for efficient deployment of LLMs. Compared to weight-only quantization, weight-activation quantization presents greater challenges due to the presence of outliers in activations. Existing methods have made significant progress by exploring mixed-precision quantization and outlier suppression. However, these methods primarily focus on optimizing the results of single matrix multiplication, neglecting the bidirectional propagation of quantization errors in LLMs. Specifically, errors accumulate vertically within the same token through layers, and diffuse horizontally across different tokens due to self-attention mechanisms. To address this issue, we introduce BiSup, a Bidirectional quantization error Suppression method. By constructing appropriate optimizable parameter spaces, BiSup utilizes a small amount of data for quantization-aware parameter-efficient fine-tuning to suppress the error vertical accumulation. Besides, BiSup employs prompt mixed-precision quantization strategy, which preserves high precision for the key-value cache of system prompts, to mitigate the error horizontal diffusion. Extensive experiments on Llama and Qwen families demonstrate that BiSup can improve performance over two state-of-the-art methods (the average WikiText2 perplexity decreases from 13.26 to 9.41 for Atom and from 14.33 to 7.85 for QuaRot under the W3A3-g128 configuration), further facilitating the practical applications of low-bit weight-activation quantization.

Hangyu Li, Hongyue Wu, Guodong Fan et al.

As privacy protection gains increasing importance, more models are being trained on edge devices and subsequently merged into the central server through Federated Learning (FL). However, current research overlooks the impact of network topology, physical distance, and data heterogeneity on edge devices, leading to issues such as increased latency and degraded model performance. To address these issues, we propose a new federated learning scheme on edge devices that called Federated Learning with Encrypted Data Sharing(FedEDS). FedEDS uses the client model and the model's stochastic layer to train the data encryptor. The data encryptor generates encrypted data and shares it with other clients. The client uses the corresponding client's stochastic layer and encrypted data to train and adjust the local model. FedEDS uses the client's local private data and encrypted shared data from other clients to train the model. This approach accelerates the convergence speed of federated learning training and mitigates the negative impact of data heterogeneity, making it suitable for application services deployed on edge devices requiring rapid convergence. Experiments results show the efficacy of FedEDS in promoting model performance.

Jianmao Xiao, Shizhan Chen, Qiang He et al.

Android utilizes a security mechanism that requires apps to request permission for accessing sensitive user data, e.g., contacts and SMSs, or certain system features, e.g., camera and Internet access. However, Android apps tend to be overprivileged, i.e., they often request more permissions than necessary. This raises the security problem of overprivilege. To alleviate the overprivilege problem, this paper proposes MPDroid, an approach that combines static analysis and collaborative filtering to identify the minimum permissions for an Android app based on its app description and API usage. Given an app, MPDroid first employs collaborative filtering to identify the initial minimum permissions for the app. Then, through static analysis, the final minimum permissions that an app really needs are identified. Finally, it evaluates the overprivilege risk by inspecting the apps extra privileges, i.e., the unnecessary permissions requested by the app. Experiments are conducted on 16,343 popular apps collected from Google Play. The results show that MPDroid outperforms the state-of-the-art approach significantly.

Lijing Zhang, Xiaowang Zhang, Zhiyong Feng

In this paper, we present an embedding-based framework (TrQuery) for recommending solutions of a SPARQL query, including approximate solutions when exact querying solutions are not available due to incompleteness or inconsistencies of real-world RDF data. Within this framework, embedding is applied to score solutions together with edit distance so that we could obtain more fine-grained recommendations than those recommendations via edit distance. For instance, graphs of two querying solutions with a similar structure can be distinguished in our proposed framework while the edit distance depending on structural difference becomes unable. To this end, we propose a novel score model built on vector space generated in embedding system to compute the similarity between an approximate subgraph matching and a whole graph matching. Finally, we evaluate our approach on large RDF datasets DBpedia and YAGO, and experimental results show that TrQuery exhibits an excellent behavior in terms of both effectiveness and efficiency.