This is the official implementation of our ICML 2024 paper "MultiMax: Sparse and Multi-Modal Attention Learning""

An Artificial Neural Network (ANN) built using PyTorch to classify images in the Fashion MNIST dataset, utilizing GPU support for improved training performance.

This repository features hands-on Jupyter Notebooks, covering everything from fundamental concepts to advanced neural network architectures.

Open ideas. Code less. Publish more.

By Ikansh Mahajan, 3C53, 102303754, COE'27

Interactive deep learning project to predict IPL innings scores from match context such as teams, venue, batsman, bowler, runs, wickets, and overs.

We introduce two novel hybrid activation functions: S3 (Sigmoid-Softsign) and its improved version S4 (Smoothed S3)

This is a custom-built neural network that detects handwritten numbers from image inputs. It uses ReLU activation in the hidden layers and a softmax activation function in the output layer for classification. The model is trained using backpropagation with a loss function to minimize prediction errors, achieving over 99% accuracy when predicting

Parametric Sigmoid Linear Unit

This is a Fake and AI image prediction using Transfer Learning

Linear and Non Linear Activation Functions : Linear, ReLU, Sigmoid, Softmax, Tanh

A Benchmark for Activation Function Exploration for Neural Architecture Search (NAS)

Graph by matplotlib

Vanishing Gradient and Activation Functions

An interactive 3D "Glass Box" sandbox built with pure NumPy and WebGL to visualize neural network math, loss trajectories, and decision boundaries in real-time.