Authors: Eduardo Alvarez, Ben Consolvo, Jack Erickson
The day before the Intel Innovation 2022 conference sessions started, Intel hosted a hackathon event with the theme of “AI for Social Good.” The three applications that participants could choose from were:
- Machine Learning: Access to a safe and sanitary water supply is essential to human life, as well as surrounding ecosystems that are experiencing the effects of droughts, pollution, and rising temperatures. Participants in this track developed and deployed a model that predicts whether freshwater is safe to drink and use for the ecosystems that rely on it.
- Computer Vision: Weeds can lead to lower agricultural yields and inefficient deployment of resources by farmers. Pesticides can be used to remove weeds, but aggressive pesticides create health risks for humans. Computer vision technology can automatically detect the presence of weeds and use targeted remediation techniques to remove them from fields with minimal environmental impact. The goal of this track was to develop and deploy a model to a drone that mitigates weeds, to maximize crop yields.
- Natural Language Processing: Humor is a critical component of personal communication, and it provides valuable information to linguistic and sociological phenomena. In this track, participants used the latest transformer-based NLP models to train and deploy a model to predict whether a statement is humorous or not. The principles learned in this track are applicable to a much wider array of NLP classification problems.
Over 100 developers, including over 30% women, participated in this full-day event. Developers from companies such as Accenture, Deloitte, HSBC, Kia, Lucid Motors, Microsoft Intune, Oracle, Twitter, US Air Force, and Vanguard participated. And students from Northeastern University, San Jose State University, Stanford University, University of California-Berkeley, University of California-Davis, and University of Manchester also competed. The nine-hour agenda of training, hacking, collaboration, and networking featured:
- A welcome from Scott Apeland, Head of Developer Ecosystem Programs at Intel, getting everyone excited for the event
- A keynote from Wei Li, Vice President and General Manager of AI & Analytics at Intel, which helped set the tone for the hackathon and communicate Intel Software efforts with oneAPI, AI Reference Kits, and Intel® AI Analytics Toolkit
- A brief training on Habana® Gaudi® hardware and software, which the teams would have access to during the hackathon, presented by Leon Goldgeisser of Habana Labs
- Access to Gaudi-based Amazon EC2 DL1 Instances for development during the hackathon
- Extensive use of the Intel Xeon 3rd generation CPUs hosted on AWS
- An overview of the AI Reference Kits, which the hackathon challenges were based on, presented by Mo Nomeli from Accenture and Louie Tsai from Intel
- An introduction to MLOps and the importance of machine learning lifecycle management in deploying performant AI models into production, by Eduardo Alvarez of Intel
- Prizes for the winning teams, which included thousands of dollars in cash as well as Intel® NUC 12 Pro X Dragon Canyon Pro workstations
The Main Event
After the introductory training sessions, teams were drafted into one of three different competition tracks. Each track had a challenge prompt meant to encompass the social good mission and emphasize the importance of models making it into production.
Throughout the afternoon, teams submitted their models into a staging environment. Once models made it into the staging environment, they were picked up by an API that automatically scored models on their F1 score (binary-classification accuracy) and inference time.
The Results
A major hardware highlight for all the participants was the fast turnaround from using Habana Gaudi-based Amazon EC2 DL1 training instances to transfer and learn the DistilBERT language model on the provided humor dataset.
At the same time, participants were pleasantly surprised by the speedup of standard AI frameworks running on CPUs, a result of Intel AI software optimizations powered by oneAPI. And many took advantage of quantizing their models to INT8 data types using Intel Neural Compressor, accelerating inference with 3rd Gen Intel® Xeon® Scalable Processors on AWS platforms.
The Grand Prize winners, who came from San Jose State University, delivered a truly innovative solution to the computer vision challenge. Their submission yielded the highest F1 score and fastest inference of the computer vision track. On top of that, their unique approach to data engineering and model optimization outshined the rest of the other track winners. Upcoming blog posts will feature the impressions from the winners, as well as more detail on their submissions.
And all the developers walked out of the hackathon with plenty of Intel gear, along with the official Innovation 2022 Hackathon shirt.
The level of participation, as well as the quality of submissions, was impressive. Intel looks forward to engaging the developer community and cultivating a strong relationship with students and innovators.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.