my implementations of machine learning papers for learning and practice

Paper-to-code replication workspace with Jupyter notebooks and Python implementations of ML and deep learning research ideas.

Paper Replications

Evidence-first quantitative paper replications with claim ceilings and machine-checkable metadata

Image-processing pipeline converting architectural drawings into binary line representations

Independent replication of the 5-minute Opening Range Breakout on QQQ (Zarattini & Aziz 2023, SSRN 4416622), stress-tested for execution costs. Break-even at ~2.2¢/share slippage; NQ confirmation filter significant per-trade (t=2.05) but 76% of its PnL is 2022 alone. Placebo control + bootstrap CIs.

Jupyter notebook workspace for replicating ML paper ideas and experiments.

Unofficial, From-scratch PyTorch replication of the Conformer paper (Gulati et al., 2020) — encoder, RNN-T decoder, training loop, and NeMo weight validation. Built block by block with documented maths.

Learning Pytorch from learnpytorch.io, implementing on datasets.

Replication of "Neural Collaborative Filtering" (He et al., WWW 2017) in PyTorch — GMF, MLP & NeuMF on MovieLens 1M + Pinterest

📈 Independent replication and out-of-sample validation of Zarattini, Aziz & Barbon (2024) "Beat the Market" intraday momentum on SPY and ES futures, with ex-ante frozen anti-overfitting protocols.

Replicating 'Reinforcement Learning from Language Feedback' (Klissarov et al., 2026) — Gemma 3 12B, GRPO, multi-turn teacher-student training on Omni-MATH

This repository contains my PyTorch practice notebooks

Hands-on replications of landmark AI papers — LeNet5, Word2Vec, Dropout & Federated Learning. From reading research to running it. Built with Python & PyTorch.

U-Net from scratch — replicating Ronneberger et al. (2015) on DRIVE retinal vessel segmentation

Replication of R-DCNN (Li et al., Eye 2023) for joint optic disc and cup segmentation