Large language models (LLMs) are increasingly used by end users, yet existing personalization methods relying on static profiles or text-only signals fail to capture query-specific expertise variation.We present ExPerT, a querywise personalization framework that adapts LLM responses to users' query domain expertise by combining semantic and behavioral cues.ExPerT consists of two key components: (i) a semantic-behavioral expertise inference module that jointly interprets query text and keystroke dynamics via in-context LLM prompting, and (ii) an expertise-conditioned response generation that adapts the level of detail, terminology, and conceptual complexity.Our user study with 40 participants and 1270 queries demonstrated that ExPerT reduced expertise inference error by 65.7% compared to the strongest baseline (MAE = 0.398 vs. 1.162) and improved response satisfaction by 17.52% (from 3.71 to 4.36) on a 5-point Likert scale....

Hallucinations in Multimodal Large Language Models (MLLMs) where generated responses fail to accurately reflect the given image pose a significant challenge to their reliability. To address this, we introduce ConVis, a novel training-free contrastive decoding method. ConVis leverages a text-to-image (T2I) generation model to semantically reconstruct the given image from hallucinated captions. By comparing the contrasting probability distributions produced by the original and reconstructed images, ConVis enables MLLMs to capture visual contrastive signals that penalize hallucination generation. Notably, this method operates purely within the decoding process, eliminating the need for additional data or model updates. Our extensive experiments on five popular benchmarks demonstrate that ConVis effectively reduces hallucinations across various MLLMs, highlighting its potential to enhance model reliability.

Recently, LiDARs have attracted a lot of attention because they enable Advanced Driver Assistance Systems (ADAS) by precisely measuring their surroundings. However, with this increased interest, adversarial attacks on LiDARs have also been demonstrated. Since LiDARs, by default, recognize the strongest reflected pulse among multiple returning pulses, these attacks exploit high-intensity laser pulses to remove original real points and create spoofed points in the point cloud. These spoofed points can serve as perturbations that cause detection errors in a object detection model using LiDAR data. To detect spoofed points, existing defense methods have only analyzed their locations in a point cloud (i.e., x, y, z). Although LiDARs provide the intensity of the reflected pulses, defense methods have ignored the intensity when detecting spoofed points. In this paper, we present a new method to detect spoofed points that are created by high-intensity laser pulse injection. Based on the fact that a higher intensity level is required to generate spoofed points in a point cloud, our method is designed to analyze the intensity of reflected pulses to identify injected pulses by attackers. Using the Adam optimization algorithm, our method resulted in a false positive rate of 3.88% for SECOND and 4.40% for PointPillar. The true positive rate was generally over 40% higher than that of other defense methods, and at least 10% higher in the worst-case scenario.

Fast Fourier transform (FFT) accelerators for IEEE 802.11ax (WiFi 6) must be designed to satisfy the stringent performance requirements specified by the standard. The design space consisting of several architectural options such as radix, memory port and loop/block pipelining is explored to optimize an FFT accelerator. Moreover, the relevant ablation study is conducted to evaluate the impact of each architectural option on the area and performance of an FFT accelerator. A commercially available high-level synthesis (HLS) tool is used to efficiently perform the design space exploration (DSE). The experimental results show that a radix-4, dual-port, fully-pipelined FFT accelerator decreases the latency by 4.7x and increases the throughput by 3x, compared to the baseline, thereby satisfying the system requirements for iterative detection and decoding (IDD) for IEEE 802.11ax. It is also shown that the proposed DSE using HLS helps to achieve a 5% improvement in area-delay product (ADP) while consuming significantly less design time than the conventional hand-coded RTL designs.