dropout

🧠 Predict COVID-19 mortality rates using a Multilayer Perceptron model, built from CDC data, without high-level frameworks.

🔍 Enhance model robustness with noise injection techniques to tackle messy, real-world data and improve machine learning performance.

Deep learning Projects with code

Addons for Kodi (XBMC)

A tiny neural network, trained live in the browser with hand-written backpropagation, fitting a noisy y = x² curve — with every weight, every bias, and every dropout mask visible as it happens, so you can watch dropout do its job instead of just reading about it.

Can language models recognize perturbations applied to their activations? Study this question via localization, classification, and in-context learning experiments.

Regularization is the constraint that forces generalization — whether by shrinking weights (L1/L2), injecting randomness (dropout, augmentation), or stopping before memorization begins (early stopping)

✈ Student 🛫 Dropout 🚁 Explainable 🚀 AI 🛸 is 🚢 an 🛥 advanced ⛴ Machine ⛵ Learning 🛼 Educational ⛱ Data ☂ Mining 🚃 Explainable 🛢Artificial 🚂 Intelligence 🕍 research 🕌 designed 🧱 to 🚈 identify 🍔 students 🍎who 🍏may 🍑 be 🍋 at 🍊risk 🫐 of 🍓 dropping 🥦 out 🫑 more 🫒 importantly 🧅why 🍔 a 🍟 model 🧵 consider ☎ a 📙 student at risk

Solutions and notes for Week 7 (Neural Networks II) of MITx 6.036, including the homework notebook, Python implementations of Linear, Tanh, ReLU and SoftMax layers, NLL loss, SGD training routines, reproducible examples, and notes on optimization, regularization, and dropout.

Predict the likelihood of a student dropping out of school

Audio crackling, stuttering, and video judder that only happen in fullscreen, only on battery or low cpu profile, and only when a real-time DSP chain (Equalizer APO, ASH, VST host, VoiceMeeter) is active.

A comparative analysis of 4 optimizer/regularization configurations (Baseline SGD, SGD+Momentum, Adam+High Dropout, AdamW+Cosine Annealing) on a PyTorch deep neural network, with live experiment tracking via Weights & Biases.

An 8-class CNN built from scratch in PyTorch to classify blood cell types, achieving 93.8% accuracy on a hidden test set, with a systematic ablation study on BatchNorm and Dropout.

Interactive Marimo WASM app for teaching neural-network capacity, overfitting, dropout, and L2 regularization.

Deep learning project that predicts wine type from chemical properties like acidity, alcohol, and sulfur dioxide. Built with Keras using BatchNormalization, Dropout, and class weights to handle imbalance. 99.77% test accuracy.

A systematic experiment on dropout placement in stacked LSTMs for stock price forecasting, with full reproducibility

Deep Learning project that predicts whether an individual earns more than $50K/year using the Adult Census dataset. Built with Keras, TensorFlow, Scikit-Learn, ANN architectures, regularization techniques, and hyperparameter tuning.

PyTorch implementation of Last-Layer MC Dropout for epistemic uncertainty estimation in Medical AI. Automatically identifies clinical edge cases and label noise in chest X-rays.

Modelo de Deep Learning para prever níveis ideais de umidade do solo e otimizar sistemas de irrigação agrícola. O projeto usa redes neurais artificiais para ajudar produtores a economizar água e energia enquanto maximizam a produtividade das colheitas.