Samya Mukherjee

Mr. Samya Mukherjee

Data Science, Ramakrishna Mission Vivekananda Educational and Research Institute, Kolkata, India
orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 July 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
Samya MukherjeeSamya Mukherjee
Samya Mukherjee
SCSayak Chowdhury
Sayak Chowdhury
Deep Reinforcement Learning (DRL) algorithms are frequently compared on standard bench mark environments, yet the practical implications of algorithmic choice for engineered, par tially observable tasks are less often examined side by side with such benchmarks. In this work we present a two-part empirical study. First, we implement ve DRL algorithms en tirely from scratch in PyTorch Deep Q-Network (DQN), Double DQN, REINFORCE, Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C) and evaluate them under matched conditions on the LunarLander-v2 control task from the Gymnasium suite. We report learning curves, sample e ciency, stability, and nal policy quality, and nd that the value-based methods (DQN, Double DQN) solve the environment reliably (in 461 and 798 episodes respectively, reaching average scores above 200), whereas the on-policy methods REINFORCE, PPO, and A2C fail to solve the task within their allotted training budgets despite requiring substantially more environment interactions. Second, we design and evaluate Snow Drone Rescue, a custom, partially observable, stochastic environment built in pygame that models an autonomous drone performing a multi-objective search-and rescue mission under battery constraints and adverse weather. A DQN agent trained with a dense-plus-sparse shaped reward learns energy-aware navigation and obstacle avoidance despite receiving only a six-dimensional partial observation. Taken together, the two stud ies indicate that, for discrete-action tasks with reasonably dense reward signals, algorithmic stability mechanisms (experience replay, target networks) matter more for nal performance than the choice between value-based and policy-gradient families, while reward and envi ronment design remain at least as consequential as the learning algorithm itself. We discuss the theoretical basis for these observations, algorithmic and environment-design trade-o s, and directions for future work, including entropy-regularized policy optimization, systematic hyperparameter search, and extension to continuous-control and multi-agent settings.

4 August 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
BMBappaditya Mukhopadhyay
Bappaditya Mukhopadhyay
Samya MukherjeeSamya Mukherjee
Samya Mukherjee
India's Pradhan Mantri Fasal Bima Yojana carries a pricing problem its design has never resolved. By capping farmer contributions at 2% of the sum insured for kharif crops, the scheme leaves private insurers no route to recover actuarial costs through the farmer-facing premium, so commercial loading and cross-state subsidies accumulate inside tender schedules that neither farmers nor state governments can audit. We model this as a rent-extraction problem under unverifiable insurer cost: absent a public benchmark, insurers' markups are bounded only by the government's political tolerance for tender prices rather than by their true underlying cost, and we show formally that publishing a verifiable actuarial reference price shifts insurers to a detection-constrained equilibrium in which rent extraction falls without any change in farmer-facing coverage. We then construct that reference price using three empirical inputs. A D'Agostino-Pearson omnibus test across 584 Indian districts establishes that kharif monsoon rainfall follows a log-normal process (DP = 2.14,p = 0.9984); the normal approximation embedded in commercial schedules underestimates extreme-event probabilities by roughly 40%. A two-way fixed-effects panel regression yields a rainfall-yield elasticity of ˆβ1 = 0.0774, well below Green Revolution-era benchmarks, reflecting four decades of groundwater expansion that has buffered yields against monsoon shortfalls. A hybrid ARIMAX–LSTM ensemble generates state-level yield forecasts with a mean absolute percentage error of 3.41% over 2021–2023, while the EWMA rainfall baseline achieves 1.40% forecast error by the final year. Feeding these inputs into a closed-form Black-Scholes put-spread formula across nine major paddy-producing states reduces required government subsidy by 41.8%–76.0% relative to current PMFBY expenditure, with no reduction in coverage for subsistence farmers. Consistent with the model, the gap between published PMFBY rates and the proposed benchmark averages 46.9% and is largest in commercially integrated states with the greatest scope for undetected rent extraction, while states with a history of high claims– where negotiated rates already track cost– show the smallest gap.