- Built a custom training engine on top of stable diffusion to generate images based off any topic
- Grid cells and their potential application in AI paper on arxiv
- reimplemented grid cells DeepMind paper
- reimplemented word2vec for a deeper understanding
- experiments with convolutions and correlations in numpy
- ConceptNet – I created a training set of generative animations to feed to a computer, a small mvp.
- Ran a hackathon and gave a talk for creatives doing machine learning with RGA
- Converted neural-style deep learning image enhancing algorithm into an adobe plugin and put it on their store.
- sketchnet – a deep learning RNN model that learns to write code to draw pictures
- built a webrtc audio/video chat service with command recognition using DeepSpeech as a prototype
- created a dataset of realistic CAD models that were used for touch simulations and real life models of thousands of objects for sensorimotor experiments
- Fix bugs and made custom fork of Intel Reinforcement Learning Coach and use for my RF experiments
- experiments with converting images into coloring books.
- Ported bindings for tensorflow to run with ruby
- a raspberry pi tensorflow autonomous driving robot called TBot
- tensorflow resources
- Built a “heroku for machine learning” platform to easily train, measure performance, and convert models into APIs. Used for internal model experimentation and external customers.
- Ran and gave talk: “ML for artist conf day” at Bocoup Boston
- Built custom deep dream models to hallucinate anything
- constructed a RNN model based off of all of Helen Keller’s books to see if I could get a model to speak in a style similar to her
- Used deep learning to parse images and automatically turn text links and emails into clickable links. The project was launched on the client’s production site processing ~5,000 images/day.
- built internal emailing software to generate leads using multi-arm bandits.
- Developed a social media software suite that crawls social media networks and generates features fed into a random forest classifier to predict who is interested in networking with you. Wrote the whole thing in 2 days. It generated several years of income with near zero maintenance.
- Seung Neuroscience Lab MIT – worked on obtaining more human created training data to feed into convolutional neural networks to classify brain tissue segments, a meta loop!
- TrueLens – developed and wrote hundreds of experiments using bayesian classifiers, SVMs, NNs, Logistic Regression, Random Forest, etc to mine and combine buying history with social data. The main goal was find out if we could use public social data to optimize offline buying habit, answer: yes.
- Socmetrics – Started a company around the idea to mine social media data. Built a people web crawler and search engine using a simplified version of PageRank and tf-idf algorithm. Raised money for this from Google Ventures and others. Got dozens of companies paying for a subscription.
- cld – ruby gem for fast language detection using Chrome language parser