About the challenge

Machine learning engineers often face a hurdle when developing models: lack of a unified platform for deep learning experimentation. Determined AI provides exactly that – a single environment for depp learning. We take care of everything you want - hyperparameter search, distributed training, experiment tracking and resource management, all in one platform. Porting your model code to one of our APIs allows you to access all these features and the ability to manage them in an easy-to-use web interface.  

In this hackathon, you are free to build any deep learning project you desire. We only require that you use the Determined platform for model development (details below).

Get started

Getting started: Intro to Determined, A First Time User’s Guide

Requirements

What to Build

Use the Determined AI platform to build a deep learning project of your choice.

What to Submit

  • Link to a GitHub public code repository containing all of the code for your project under a permissive OSI-approved license, such as Apache License 2.0 or MIT.

  • Your final model uploaded to Hugging Face Model Hub 

  • Link to the dataset used on Hugging Face Datasets 

  • A README.md which includes: 

  • Your objective,  

  • A data sample from your dataset with an explanation,  

  • A description of your model architecture, 

  • Instructions for how to run your training job, 

  • A screenshot of your best metrics from your WebUI, 

  • A description of your evaluation metrics, and

  • Your evaluation results given these metrics.

  • A quick 2-3 minute video summary explaining the project, showing the experiment metrics and evaluation results of both the trained model and the untrained model, uploaded to YouTube.

Hackathon Sponsors

Prizes

$10,000 in prizes
Best Medical Application
1 winner

Best project that investigates a dataset or builds a model pertaining to a medical application, such as biotech, human disease research, etc.

Best NLP application
1 winner

Best project that trains a model with text as inputs/outputs.

Best Computer Vision Project
1 winner

Best project that trains a model with images or videos as inputs/outputs.

Best Multimodal Project
1 winner

Best project that trains a model with inputs/outputs of multiple modalities. The valid modalities are: vision, language, audio.

Creativity
1 winner

This prize will be awarded to the project showing an ML use case that is not commonly seen or one that solves an "AI for Good" problem (e.g. reducing the impact of climate change).

Devpost Achievements

Submitting to this hackathon could earn you:

Judges

Hoang Phan
HPE

Judging Criteria

  • Technical quality
    How well was the idea implemented technically?
  • Utilization of Determined
    To what extent were Determined features utilized?
  • Creativity
    How original is the idea?

Questions? Email the hackathon manager

Tell your friends

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.