Steering Vector (SV) is a powerful technique for controlling Large Language Models (LLMs) by manipulating their activations without altering model weights. However, when constructed from sensitive data, SV poses significant privacy risks, as it may leak private information. Existing differential privacy (DP) techniques for constructing SV cannot be directly applied to training-based SV construction paradigms, which offer higher task performance. In this work, we present **PrivSV**, a general privacy-preserving approach for constructing SV with DP guarantees, compatible with arbitrary SV construction paradigms while maintaining high utility. In PrivSV, we propose three novel methods: a Layer-wise Noise-Resilient Reduction (LNR²) method to reduce the injected noise in high-dimensional SV; a Directional Prior Compensation (DPC) method to recover utility degraded by noise perturbation; and a Privacy-Aware Optimal Parameter Determination (POPD) method to adaptively maximize the performance of the final compensated SV. Extensive experiments on open-source LLMs of different families (i.e., LlaMa, Qwen, Mistral and Gemma) demonstrate that PrivSV outperforms several existing techniques across various privacy budgets.

Randomized smoothing (RS) is popular for providing certified robustness guarantees against adversarial attacks. The average certified radius (ACR) has emerged as a widely used metric for tracking progress in RS. However, in this work, for the first time we show that ACR is a poor metric for evaluating robustness guarantees provided by RS. We theoretically prove not only that a trivial classifier can have arbitrarily large ACR, but also that ACR is extremely sensitive to improvements on easy samples. In addition, the comparison using ACR has a strong dependence on the certification budget. Empirically, we confirm that existing training strategies, though improving ACR, reduce the model’s robustness on hard samples consistently. To strengthen our findings, we propose strategies, including explicitly discarding hard samples, reweighing the dataset with approximate certified radius, and extreme optimization for easy samples, to replicate the progress in RS training and even achieve the state-of-the-art ACR on CIFAR-10, without training for robustness on the full data distribution. Overall, our results suggest that ACR has introduced a strong undesired bias to the field, and its application should be discontinued in RS. Finally, we suggest using the empirical distribution of $p_A$, the accuracy of the base model on noisy data, as an alternative metric for RS.

We propose road-segment anything model (SAM), a universal model for extracting roads from large, very-high-resolution (VHR), optical, remote sensing (RS) images. Unlike previous methods, Road-SAM builds upon the foundation of the SAM, a large-scale image-segmentation model, to explore a new paradigm for customizable road extraction (RE). Within the framework, we introduce three variants that allow for flexible insertion of adapters at different positions within the transformer block. Additionally, the model employs a task-specific input module of explicit visual prompting (EVP) during training that uses embedded features and high-frequency component (HFC) information as prompts. Road-SAM also utilizes a carefully designed frequency adapter fine-tuning mechanism, leveraging lightweight yet effective fine-tuning techniques to integrate domain-specific RS knowledge into the RE model, enhancing segmentation performance and making efficient use of computational resources. Comprehensive experiments on two sets of RE benchmark datasets demonstrate the effectiveness of the proposed method. Extensive ablation experiments further validate its superiority over multiple state-of-the-art (SOTA) RS RE algorithms, with updates applied to only 10% of the parameters.

MOTIVATION: A critical issue in drug benefit-risk assessment is to determine the frequency of side effects, which is performed by randomized controlled trails. Computationally predicted frequencies of drug side effects can be used to effectively guide the randomized controlled trails. However, it is more challenging to predict drug side effect frequencies, and thus only a few studies cope with this problem. RESULTS: In this work, we propose a neighborhood-regularization method (NRFSE) that leverages multiview data on drugs and side effects to predict the frequency of side effects. First, we adopt a class-weighted non-negative matrix factorization to decompose the drug-side effect frequency matrix, in which Gaussian likelihood is used to model unknown drug-side effect pairs. Second, we design a multiview neighborhood regularization to integrate three drug attributes and two side effect attributes, respectively, which makes most similar drugs and most similar side effects have similar latent signatures. The regularization can adaptively determine the weights of different attributes. We conduct extensive experiments on one benchmark dataset, and NRFSE improves the prediction performance compared with five state-of-the-art approaches. Independent test set of post-marketing side effects further validate the effectiveness of NRFSE. AVAILABILITY AND IMPLEMENTATION: Source code and datasets are available at https://github.com/linwang1982/NRFSE or https://codeocean.com/capsule/4741497/tree/v1.

Reconfigurable intelligent surfaces (RISs) are able to provide passive beamforming gain via low-cost reflecting elements and hence improve wireless link quality. This work considers two-way passive beamforming design in RIS-aided frequency division duplexing (FDD) systems where the RIS reflection coefficients are the same for downlink and uplink and should be optimized for both directions simultaneously. We formulate a joint optimization of the transmit/receive beamformers at the base station (BS) and the RIS reflection coefficients. The objective is to maximize the weighted sum of the downlink and uplink rates, where the weighting parameter is adjustable to obtain different achievable downlink-uplink rate pairs. We develop an efficient manifold optimization algorithm to obtain a stationary solution. For comparison, we also introduce two heuristic designs based on one-way optimization, namely, time-sharing and phase-averaging. Simulation results show that the proposed manifold-based two-way optimization design significantly enlarges the achievable downlink-uplink rate region compared with the two heuristic designs. It is also shown that phase-averaging is superior to timesharing when the number of RIS elements is large.