SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models

Safety-aligned language models often exhibit fragile and imbalanced safety mechanisms, increasing the likelihood of generating unsafe content. In addition, incorporating new knowledge through editing techniques to language models can further compromise safety. To address these issues, we propose SafeInfer, a context-adaptive, decoding-time safety alignment strategy for generating safe responses to user queries. SafeInfer comprises two phases: the safety amplification phase, which employs safe demonstration examples to adjust the model's hidden states and increase the likelihood of safer outputs, and the safety-guided decoding phase, which influences token selection based on safety-optimized distributions, ensuring the generated content complies with ethical guidelines. Further, we present HarmEval, a novel benchmark for extensive safety evaluations, designed to address potential misuse scenarios in accordance with the policies of leading AI tech giants.

Dynamic Ego-Velocity estimation Using Moving mmWave Radar: A Phase-Based Approach

Precise ego-motion measurement is crucial for various applications, including robotics, augmented reality, and autonomous navigation. In this poster, we propose mmPhase, an odometry framework based on single-chip millimetre-wave (mmWave) radar for robust ego-motion estimation in mobile platforms without requiring additional modalities like the visual, wheel, or inertial odometry. mmPhase leverages a phase-based velocity estimation approach to overcome the limitations of conventional doppler resolution. For real-world evaluations of mmPhase we have developed an ego-vehicle prototype. Compared to the state-of-the-art baselines, mmPhase shows superior performance in ego-velocity estimation.