11 papers
Enforcing LLM Safety through DMD-based Classification of Prompt-Response Embedding Dynamics
Mohamed Akrout, Olivera Kotevska, Dan Wilson
Large Language Models (LLMs) are increasingly deployed in high-stakes applications, yet their tendency to generate toxic, harmful, or policy-violating content poses significant ris…
Optimality of Bang-Bang Switching for Breaking the Chu Limit via Time-Modulated Matching
Mohamed Akrout, Miguel Rodrigo Castellanos
This paper shows that surpassing the Chu limit on -factors via time-modulated matching requires non-smooth switching strategies. We derive a nonlinear differential condition to…
Guarantees on Dynamical System Distinguishability for LLM Token Generation
Mohamed Akrout, Dan Wilson
Recent work has shown that classifying large language models (LLMs)' responses can be distinguished by modeling token embeddings as trajectories of a black-box dynamical system (DS…
Time-varying Wireless Channel Tracking with Online Parameter Learning via the Birth-Death-Drift Model
Tiancheng Gao, Mohamed Akrout, Faouzi Bellili +1
Accurate massive MIMO channel state information (CSI) acquisition with low pilot overhead is critical in dynamic propagation environments. Exploiting temporal correlation is key to…
Low-Cost Black-Box Detection of LLM Hallucinations via Dynamical System Prediction
Dan Wilson, Mohamed Akrout
Large Language Models (LLMs) frequently generate plausible but non-factual content, a phenomenon known as hallucination. While existing detection methods typically rely on computat…
Signal Processing Foundations of Reconfigurable Antennas in the Tri-Hybrid MIMO Architecture
Nitish Vikas Deshpande, Joseph Carlson, Siyun Yang +6
To enable larger apertures in multipleinput multipleoutput MIMO systems the trihybrid MIMO architecture offers a promising lowcost and lowpower solution by introducing reconfigurab…