gradient-clipping

Official implementation of the paper: "ZClip: Adaptive Spike Mitigation for LLM Pre-Training".

Object Classification Training Framework

[Re] Can gradient clipping mitigate label noise? (ML Reproducibility Challenge 2020)

Code for the NeurIPS 2020 paper "Improved analysis of clippind algorithms for non-convex optimization", including various clipping algorithms for neural network training in CV and NLP.

Global L2 norm adaptive gradient clipping engine to mitigate exploding gradients during deep neural network training.

Global L2 norm adaptive gradient clipping engine to mitigate exploding gradients during deep neural network training.

High-performance, branchless numerical stability kernels and compiler-optimized core infrastructure for advanced JAX/XLA deep learning architectures.

Federated Learning IDS — Privacy Attack Analysis

PyTorch extension for alternative backward rules and gradient transforms (STE, gradient jamming, non-standard activations).

Finsler-Adam: Asymmetric-metric optimizer with critical scaling gradient clipping for PyTorch. Drop-in AdamW replacement.

NetGuard is a production-grade Network Intrusion Detection System (NIDS) designed to classify sophisticated cyber threats across 15 distinct attack vectors. By treating network traffic as both a continuous 1D spatial signal and a discrete temporal sequence, NetGuard bypasses the limitations of single-modality intrusion detection.

Neural network for machine translation using sequence to sequence model on a dataset of English and French sentences

Coursera - RNN Programming Assignment: In this project, we are in charge of a special task: Leading biology researchers are creating new breeds of dinosaurs and bringing them to life on earth, and our job is to give names to these dinosaurs. If a dinosaur does not like its name, it might go berserk, so choose wisely!

StableStockPredictor is a robust deep learning model for predicting S&P 500 stock prices, built with TensorFlow and Keras. It leverages LSTM networks with gradient clipping, robust scaling, and stable feature engineering (e.g., RSI, moving averages, volatility) to ensure reliable performance in volatile markets.

Successfully established an LSTM model to effectively forecast global equity based on over 20+ years of historical data of global equity.

Text generation from Coursera deeplearning.ai "Sequence Models" course.

Recurrent neural network for building a character-level language model and its application to generating new dinosaur names

Allows automatic gradient norm clipping. This feature can help to stabilize training in certain situations by limiting the magnitude of gradient updates. The implementation is inspired by the paper "AutoClip: Adaptive Gradient Clipping for Source Separation Networks"

Predicting whether two persons meeting in a speed dating event get matched using L2 regularised simple neural network with 90% accuracy.

Build a Multi-turn Conversations Chit-Chat Bot