适用于高性能系统的多进程解压缩软件(A multiprocess decompression software for high-performance system)

😴 DeepSleep2 is a compact U-Net-inspired convolutional neural network with 740,551 parameters, designed to predict non-apnea sleep arousals from full-length multi-channel polysomnographic recordings at 5-millisecond resolution. Achieves similar performance to DeepSleep with lower computational cost.

[RLC2025] - Official repository of "SPEQ: Offline Stabilization Phases for Efficient Q-Learning in High Update-To-Data Reinforcement Learning"

▶ Advanced computer science problem sets from Harvard's CS50X, utilizing C, Python, and SQL to explore algorithmic fundamentals and web development.

▶ Full-stack web development projects for Harvard's CS50W, utilizing Django, JavaScript, SQL, and CSS to architect scalable and interactive web applications.

Pattern-aware optimization framework achieving 93.8% complexity reduction in LLM generation with <1% overhead

Boost Python's performance using Cython – a bridge between Python's simplicity and C's efficiency. Explore and learn how Cython accelerates code execution.

▶ Comprehensive data structures and algorithms roadmap in Java, providing optimal architectural implementations for arrays, linked lists, queues, trees, and graphs.

Measure the time for large-scale operations and contribute to the exploration of computational efficiency.

Split a single neural network into multiple smaller networks using weight splitting.

Hierarchy-Exploiting Ensembles for Improved Multi-Class Classification

Machine Learning Research to Advance Simulation Science

Исследование производительности операций с данными и статистический анализ товаров бисера. Сравнение медленных и оптимизированных методов обработки данных, проверка статистических гипотез и выявление закономерностей.

▶ Comprehensive Python implementations for Harvard's CS50P, covering standard library integrations, unit testing frameworks, and advanced OOP concepts.

A responsive intent recognition framework with recursive optimization that achieves high accuracy with minimal computational resources through mathematical optimization.

Frictionless Computing

エビデンスに基づくモデル選択：線形モデルが勾配ブースティング決定木を上回る状況とその理由

▶ Comparative benchmarking repository of fundamental sorting algorithms including Bubble, Selection, Quick, and Merge Sort in both Java and Python.

Official repository of "SOPE: Stabilizing Off-Policy Evaluation for Online RL with Prior Data"

A novel Transformer with Adaptive Computation and Quantized Attention.