Sham Satish Thakare

Sham Satish Thakare โ€” Publications

(He/Him/His)

Independent Computer Science Researcher

Focus: Foundation Model Reasoning & Reliable Distributed Systems


APCOER Pune (SPPU) ยท First Class with Distinction

Publications:

You can view my research repositories and preprint records at [GitHub] or [SSRN].

Manuscripts Under Review / In Submission:

Sham Satish Thakare
Under Review at IEEE Transactions on Artificial Intelligence (IEEE TAI), 2026
Falsifies token-predictive entropy as an uncertainty proxy in LM RLVR. Demonstrates that predictive entropy correlates strongly with sequence length ($r = +0.486$), causing uncertainty-weighted GRPO to systematically misallocate credit on valid multi-step derivations.
Sham Satish Thakare
Working Paper, 2026 • Distributed Systems Research
Formulates failure-domain aware latency profiling algorithms in C++20 Raft consensus clusters (quorumshift), eliminating tail latency spikes under asymmetric network fault injection and non-uniform round-trip delays.
Sham Satish Thakare
In Peer Review at Intelligence-Based Medicine (Elsevier) • SSRN Preprint 7446543
Presents layered runtime boundary enforcement and deterministic state-interception architectures for autonomous LLM tool executions in high-stakes healthcare-commerce workflows.
Sham Satish Thakare
Under Review at 11th IEEE Special Session on Machine Learning on Big Data (IEEE BigData 2026)
Formulates state-contrast continuation evaluation ($D_{\text{recovery}} = -0.1100$) measuring error recovery under frozen structural covariates across reasoning model rollouts, preventing confounding between generative capacity and recovery resilience.

Preprints & Working Papers:

Sham Satish Thakare
Submitted to Elsevier Artificial Intelligence (AIJ), 2026
Evaluates neural reasoning continuation state matching under verifier perturbations and structural covariate controls, establishing empirical bounds on reasoning recovery dynamics.
Sham Satish Thakare
Working Paper, 2026
Theoretical and empirical characterization of compute Pareto frontiers comparing test-time guided search versus amortized post-training reinforcement learning on complex reasoning graphs.
Sham Satish Thakare
Working Paper, 2026
Diagnoses parameter plasticity and representation collapse across model checkpoints, establishing diagnostic probes for RL fine-tuning stability.
Sham Satish Thakare
Working Paper, 2026
Zero-knowledge memory attestation proofs and adaptive Path ORAM tree rebalancing for hardware-enclave environments under speculative and memory access attacks.
Sham Satish Thakare
Working Paper, 2026
Formulates topological causal discovery over distributed tracing graphs, isolating anomalous propagation paths with statistical guarantees.

Blog Posts: