Publications

arXiv:2609.19499 — For self-consistency, candidate count N is not enough. At fixed N, how samples are batched can change LLM test-time scaling energy and latency by several times on A100 GPUs.

ATEST (RATTS) is resource-aware self-consistency for LLM inference. It adapts how many samples to draw to live GPU load. The same four-sample query can cost 4.1× more energy when the GPU is busy. arXiv:2610.25519.

Interpretable adaptive sampling for self-consistency. A fuzzy controller uses prompt complexity and model confidence to choose how many self-consistency samples each prompt receives.

A dependency-aware auto-scaling framework for serverless computing using multi-expert consensus, workload forecasting, ensemble learning, and cost-aware resource allocation.

PerfMamba analyzes the runtime behavior, resource utilization, memory access patterns, scaling properties, and pruning opportunities of Mamba-1 and Mamba-2 selective state space models.

A fully automated LLM-driven AutoML framework for cross-lingual handwritten OCR using closed-loop neural architecture search with GPT-5, GPT-4o, and Claude Sonnet 4 across Arabic, English, and Persian scripts.