Research, deep-dives
Nine papers, one through-line.
Long-form notes on each of the peer-reviewed papers, with abstracts, FAQs, and citations. The work spans quantum machine learning, post-quantum IoT, blockchain consensus, reinforcement learning under partial observability, AI-optimised VLSI and lightweight standards, but the instinct is the same in each: identify the place where conventional wisdom is paying a hidden cost, then redesign the contract instead of micro-optimising around it.
Hardware-Agnostic Quantum Kernel Feature Mapping for Anomaly Detection in Critical Infrastructure
8-qubit ZZFeatureMap kernels validated on IBM's 156-qubit ibm_fez. The first hardware-validated quantum-ML pipeline for industrial control system anomaly detection.
AI-Optimized VLSI Architecture for Energy-Efficient and Sustainable IoT Systems
RL, GAs, and Bayesian optimisation steering Cadence and Synopsys synthesis flows. 43.3% power, 29.7% delay, 52% energy reduction over conventional VLSI.
AI-Driven Green Cognitive Radio for Sustainable Spectrum Management
Deep RL over partial-observable spectrum, with energy-aware reward shaping that turns spectrum efficiency and power efficiency into the same optimisation problem.
Mitigating Tails Switching in Multibranch Proof-of-Stake Systems: A Quantum-Inspired Approach
Borrowing wavefunction collapse to fix proof-of-stake. Direct measure functions plus single-branch nodes that stabilise multibranch chain selection.
Quantum-Resilient IoT Energy Metering on Blockchain
Hybrid PQC plus AI plus blockchain smart-meter architecture. Sub-fifty-dollar nodes, 1850 TPS, 97.8% F1 anomaly detection, harvest-now-decrypt-later resistant.
Bootstrap Method for Microcanonical Ensembles in Quantum Systems
Lifting the bootstrap technique from classical quantum mechanics into the microcanonical ensemble. Moment-positivity constraints that tame the unphysical region.
Optimizing Reinforcement Learning in Partially Observable Environments Using Compressed Suffix Memory
Heuristic-bounded, Boltzmann-sampled successor to USM that learns faster on POMDPs without the exponential state-space explosion.
Optimizing LSM Tree Operations with Deferred Updates
Drop the hidden read penalty from secondary index maintenance. Up to 10x speedups on write-heavy workloads in RocksDB- and Cassandra-style engines.
Efficient OneM2M Standard Implementation for Lightweight IoT
Clean-sheet C implementation of oneM2M v5.1 for resource-constrained gateways, peer-to-peer devices, and cloud scenarios. Small enough to fit, fast enough to matter.