Implements a PPO agent utilizing a Graph Neural Network (GNN) for state embeddings and continuous action spaces. The policy update integrates a custom stability loss based on node features and an entropy regularization term, ensuring efficient computation and stable training.
Fine-tune GPT-2 on JSONL datasets (supporting both generic text and Q&A formats) using Hugging Face Transformers, with a focus on memory-efficient training strategies like mixed precision and gradient accumulation.
Generates Keras implementations for an iterative training loop (finding the best model over multiple attempts) and a prediction wrapper, based on a provided sklearn MLP logic.
Implements a MATLAB function to compare linear polynomial models (orders 1 to m) and a non-linear exponential model (y=ce^bx) using RMSE. Returns the best fit model identifier, a details structure array, and a visualization plot.
Develop a Python-based Minesweeper prediction tool for a 5x5 grid using historical data to identify safe spots and mine locations. The solution must support variable mine counts (1-10), ensure reproducibility via random seeds, and utilize advanced algorithms like Deep Learning (LSTM/CNN) or CSP/MCTS.
Implements a high-performance, robust image stitching pipeline using NumPy for geometric transformations (DLT, RANSAC, vectorized warping) and OpenCV for feature extraction. Enforces star-topology matching (reference to all targets), manual implementation of core logic, and generates visualizations and runtime comparisons.
Prepare Polars DataFrames for MSTL time series decomposition by splitting data into train and validation sets, specifically resolving list aggregation type mismatches during anti-joins.
Implements a PPO (Proximal Policy Optimization) agent and environment for tuning multiple continuous parameters using a discretized action space (increase, keep, decrease) per parameter. The policy network outputs a probability distribution matrix, and the environment handles parameter updates to avoid redundancy.
Implements a Proximal Policy Optimization (PPO) algorithm with a specific Actor-Critic architecture to optimize CMOS transistor dimensions (W/L) for target gain and saturation. Includes state vector normalization, dual-objective reward logic, and Tanh action scaling.
Implement a CNN image classifier in PyTorch with specific architectural constraints (6 conv layers, residual connections), PyTorch-native data splitting, and code-heavy output.
Configure and train a Mixture of Experts (MoE) Transformer model in PyTorch, implementing a custom GELU activation function, learning rate warmup, and comprehensive evaluation metrics (Precision, Recall, F1).
Provides a complete end-to-end workflow for text classification using a PyTorch Transformer model. It includes automatic vocabulary generation from raw text, a custom tokenizer implementation, data padding, model training on CPU, and visualization of loss and accuracy metrics.
Generate complete R code for hierarchical Bayesian models using Gibbs/Metropolis sampling, strictly adhering to a user-provided template that includes initialization, sampling, convergence diagnostics (trace/ACF), multi-chain execution, thinning, and chain combination.
Implements a reinforcement learning reward function for analog circuit design that handles mixed minimization/maximization objectives using normalized differences, while incorporating transistor saturation tracking to ensure stability.
Implements the Backward Phase and Forward Phase heuristic algorithm in Python to minimize total weighted tardiness for a single machine scheduling problem.
Configure and execute a time series forecasting pipeline using Polars for data manipulation and MLForecast with LightGBM for modeling, applying specific lag features, rolling statistics, and evaluation metrics.
Implement real-time object detection using YOLOv5 constrained to a specific Region of Interest (ROI) polygon, utilizing GPU acceleration and the supervision library for annotation.
Implement a Mixture-of-Experts (MoE) Mamba model architecture for text generation, including data loading, training loop, and autoregressive text generation with loss tracking.
Implements a robust panorama stitching pipeline using NumPy for geometric transformations (RANSAC, DLT, Warping) while comparing SIFT, SURF, and ORB features. Enforces reference-to-target matching and generates visualization/performance metrics.
Optimizes memory consumption during PyTorch model training by implementing mixed precision training, gradient accumulation, and efficient data loading strategies to fit within hardware constraints.
Implement a PyTorch script to generate synthetic linear equation data (ax + b = c), train and compare Mixture of Experts (LSTM and Transformer) against Single General Models (LSTM and Transformer), and visualize the training loss comparison.
Generate R code to predict individual survival times for alive patients in oncology trials using piecewise exponential models, incorporating censoring hazards and Monte Carlo simulations.
Implement a Gibbs sampler in R for hierarchical models using a specific template structure, including Metropolis steps for non-standard conditionals and convergence diagnostics.
