Translating machine learning research into production-ready software

HiddenLayer

Extract from black-boxed models and adversarially attack various architectures using replicated models

Convert pictures of food to fancy plated versions through a novel transformer architecture and latent diffusion models

Efficiently train and deploy machine learning models with limited memory

Train a model to generate audio in your own voice, using signal processing techniques, GANs and autoencoders

Create a web app where users can generate their own lofi and game beats to study/chill to using transformers and LSTMs

Attune

Use different unsupervised/clustering models to detect anomalies in high-dimensional sensor data

Train an imitation learning agent to recreate stylized motions on simulated characters

Ditto

Transfer styles of pokemon sprites onto pet images using GANs and contrastive learning

GeoKnowr

Create Geoguessr bot that predicts GPS locations given image using CNNs and LSTMs

Hummus

Humanize computer music to match the lifelikeness of real performance with VAEs and transformers

Launchkart

Develop an intelligent Mario Kart agent that can replicate the skill and technique of a human player using behavior cloning algorithms

Lilspeare

Converting rap lyrics to Shakespearian-style text using Transformers and RNNs. The 1500's never looked so dope.

Reconstructor

Creating high-quality 3D reconstructions from photos of our own. Each member will choose variable scenes and items to reconstruct with NeRF and various follow up methods.

Duet

Experiment with a VQ-VAE based network for converting audio to the style of another audio file

Inflection

Create a voice conversion model framework for voice assistants, implementing data generation, spectral analysis, and an Amazon Lex-based web app

Lightspeed

Artificially stain various cellular samples using cutting edge style transfer techniques

Lucinator

Build a production ready virtual assistant chatbot with robust machine learning pipelines and slack integration

Phodexr

Enable semantic search through images and other mediums in a zero-shot manner with CLIP

Scribe

Few-shot meta learning to learn handwriting styles and generate similar handwritten text using GANs

AI Brain

Experiment on multi-agent reinforcement learning for playing the popular sport, soccer

Arturo

Apply Computer Vision techniques to improve tiny object detection on measuring roof debris and damage

Rosetta

Build an encoder for images that can perform few-shot training across CV tasks using unsupervised learning

Shallowsleep

Explore adversarial models to track sleep stages by comparing results from watch and radio frequency data to brain wave data

Strategist

Develop a profitable trading strategy for the market today based on common technical indicators

Raise

Create an AI web app using Monte Carlo sampling, deep and reinforcement learning to play heads up and nine handed no-limit Texas Hold'em poker

Illustrator

A pictionary-like web game with a bot to generate and classify sketches

Interact

Building models that detect, contextualize, and identify relationships between humans and objects in images.

Memories

A web app where users can upload pictures from their photo album and generate ambient sounds and songs from the image’s scenery and emotion

Watchman

Merging information between multiple cameras to track objects of interest accurately in 3D spaces

Brawl

Training state of the art reinforcement learning agents to play the popular game, Super Smash Bros Melee

BackyardIOT

Animal information and tracking using a CNN with Raspberry Pi to send information to the cloud

Emergence

Reinforcement learning in competitive environments for simulating intelligence

Evolve

Universal codebook for image compression using evolutionary-based optimization methods

Inspector

Fake news detecting Chrome extension incorporating NLP techniques and stance detections

Orchestrator

Converting music between different styles/instrumentations using encoder-decoder networks in PyTorch

Tonal

Post-workout image generation using generative adversarial networks

BalliNN

NBA playoff performance prediction from regular season data using statistical learning and neural networks

ClairVoyance

Weather forecasting using high-resolution video frame prediction using CNNs and RNNs

Musecage

Answering visual questions using image and word embeddings and recurrent attention algorithms

Snapshot

Beautifying Berkeley with a #filter created with CycleGAN, an image to image translation neural network

SpeculateAI

Stock market trend prediction using NLP and time series analysis

DarkLight

Improving existing convolutional neural network architecture for real-time low-light image inferencing

DreamRL

Generative models for reinforcement learning to attain high performance with simplicity

Explorer

Analyzing exploration and curiosity-based techniques in deep reinforcement learning

GrubGen

Food image generation using generative adversarial networks

Dangersense

Detecting and responding to dangerous driving situation by predicting collisions and road closures

Ford

Predicting driver behavior to improve rider experience in autonomous vehicles using kinematics and LSTMs

MOR

Multi-Objective Robotics solution for efficient and adaptive robot operation

MeTal

Meta learning and multi-task learning

Facelift

Augmenting low-fidelity images of 2D faces using 3D facial reconstruction

Jazz

Generating jazz melodies in specified styles using Markov chains and clustering

DeepBeat

Polyphonic music generation using deep neural networks

Sherlock

Building a human-centric computer vision API that provides optimized algorithms for object detection and tracking

Stella

Voice controlled computers using natural language processing