01. Machine Learning

Four classifiers, no servers.

Every model on this page runs entirely in your browser. No API calls, no backend compute. Training happened once offline; inference is yours to run as many times as you like.

NLP & Computer Vision · University of Illinois Urbana-Champaign

Live Demos

Try the classifiers

01

N-gram Review Classifier

Paste a hotel review and compare its perplexity under two language models trained on truthful versus deceptive reviews. The model with the lower perplexity wins.

N-gram LM Laplace smoothing Perplexity Opinion spam

02

Email Spam Classifier

Paste any email and a Naive Bayes classifier trained on 10,000 Enron emails scores it as spam or ham. Model weights are pre-computed and loaded as JSON at startup.

Naive Bayes Bag of words Enron corpus Log-likelihood

03

Sentiment Classifier

A feedforward neural network trained on 8,000 Yelp reviews predicts the star rating (1 to 5) of any review. Model weights are loaded as JSON; inference runs as TypeScript with no WebAssembly required.

FFNN Bag of words Yelp reviews 5-class

04

Image Captioning

Upload an image and a Vision Transformer generates a natural-language caption in your browser. Original assignment trained VGG16 and LSTM on MS-COCO. First load downloads about 100 MB.

ViT-GPT2 Transformers.js VGG16+LSTM MS-COCO