The challenge is open for registration and submission! You can sign up for updates here. 

Executive Summary

A startup is looking to contract individuals and teams who use AI and machine learning to predict the aftermath of disruptive events - and this challenge is your pathway in. 

Top performers will be offered a paid contract to continue this work directly with the startup , with total compensation exceeding $10,000 USD. The top winner also receives a $7,000 USD cash prize.

They are looking for three things:

We are looking for trainable algorithms that predict material fragmentation and displacement resulting from a single point disruption. The ideal algorithm would improve the ability to locate materials of different sizes after a disruptive event. Examples of these types of events include asteroid collisions, building collapse, volcanoes, or landslides.

If you are currently modeling physics-driven events using AI/ML, we invite you to participate in this Challenge. 

To enter the Boom Challenge, you will need to describe your team and your physics-driven AI/ML algorithm, train your algorithm to make predictions on fragment distribution in a simulated scenario. For additional points, create an inverse design where you propose impact parameters, then upload your final submission and a video describing your algorithm.  

Eligibility Requirements 

How to Enter

For complete information, see the Submission Requirements on the Guidelines page.

The Challenge - Modeling Asteroid Impact Debris Fields

For the Challenge, we have created an imaginary stellar system called Mox-95. Like all planets, the planets within Mox-95 are subject to asteroid impacts. When asteroids strike planetary surfaces, they generate massive ejecta blankets – debris fields of fragmented rock that spray outward from the impact site. Understanding the size distribution of the debris fragments and how far they travel is critical for interpreting ancient craters and predicting hazards from future impacts – on the planets within Mox-95 as well as our own planet.

The Mox-95 stellar system experiences unusual gravitational disturbances that slightly alter the impact dynamics compared to Earth. However, the underlying physics remains self-consistent and intuitive, and many physical principles from our own solar system still apply.

Challenge Description

The Challenge has two parts: 

Forward Prediction

A Training Dataset has been compiled, made up of thousands of simulated asteroid impact events in Mox-95. The dataset includes both the impact parameters (as input) and the resulting ejecta outcomes (as output). Use this dataset to train your AI/ML algorithm to predict ejecta outcomes given impact scenarios. Once your algorithm is trained, run the Test Dataset through your algorithm to generate the ejecta output data. The Test Dataset contains out-of-distribution impact scenarios (though the physics remains the same) to test your model’s generalizability. Note: Physics informed methods are strongly encouraged.

Inverse Design

Based on what you and your model learned from the first part of the challenge, propose 20 impact scenarios that will result in ejecta outcomes satisfying the following constraints:

Included in the repository is a configuration file describing these outcome constraints and a set of input bounds. The parameters of your proposed impact scenarios should lie within the input bounds.

Since asteroid impacts are stochastic, a given impact scenario produces a distribution of possible ejecta outcomes rather than a single result. Each of your scenarios will be evaluated by its average ejecta outcome.

In addition, each scenario that satisfies the constraints will receive a “small-impact score” calculated from the impact energy and the average R95 outcome. The lower the energy and ejecta range, the higher the small-impact score. See “Scoring Metrics” in the Guidelines tab for more information.

Data Repository

The data repository contains all the information needed to complete the challenge. You can access the repository here:  https://github.com/poweredbyfreelancer/Boom-Challenge-Datasets

The repository contains:

Contact  

Please submit your questions in the challenge Clarification Board or submit them via challenges@freelancer.com .