Neurotechnology Researchers Win Kaggle Competition with Deep Neural Network Solution for The Nature Conservancy Fisheries Monitoring
Research engineers from Neurotechnology teamed up and came in first place out of 2,293 teams who entered the Kaggle competition, which tasked top scientists and algorithm developers to design an AI fish detection and classification algorithm.
Vilnius, Lithuania – June 14, 2017 – A team of deep neural network researchers from Neurotechnology won first place in a Kaggle competition that sought cutting-edge AI solutions for fisheries monitoring. For their winning solution in The Nature Conservancy Fisheries Monitoring competition, the team received the first place $50,000 prize of the $150,000 prize pool.
Founded in 2010, Kaggle is a learning, sharing and development site for data, code, research and process. One of its primary features is that of being a platform that supports predictive modeling and analytics competitions. Organizations, companies, researchers and other interested parties submit challenges in the form of competitions, posting their data and project parameters to the Kaggle site. Coders, engineers and data miners from around the world compete to produce the best solutions. Currently Kaggle has more than a million registered users. This crowd-sourcing and sharing community spans 194 countries and is one of the largest and most diverse data communities in the world.
The Fisheries Monitoring competition was one of the biggest Kaggle competitions. According to The Nature Conservancy, which initiated this competition, illegal, unreported and unregulated fishing practices are threatening marine ecosystems, global seafood supplies and local livelihoods. By using computer technology to aid in monitoring fisheries, human capital can be re-allocated to management and enforcement, helping local, regional and global partners preserve the integrity and viability of these fisheries today and into the future.
2,293 teams submitted algorithms for the identification of fish and other marine species from video streams. Competing solutions were evaluated based on an unseen test set that resembles a real-life scenario.
The Neurotechnology employees, who entered the competition independently under the team name "Towards Robust-Optimal Learning of Learning," used state-of-the-art deep neural networks to solve the problem and provide the best overall solution in the competition. The winning team is comprised of Gediminas Peksys, Ignas Namajunas and Jonas Bialopetravicius, all of whom work on Neurotechnology's AI development team, which designs and delivers a range of products and services based on deep neural networks, including computer vision and object recognition.
"This was one of the first Kaggle competitions that was comprised of two stages, which means that models developed during the first stage were frozen and evaluated on unseen data that was made available during the second stage," said Gediminas Peksys from the Towards Robust Optimal Learning of Learning team. "In such a setting, it is very easy for a team's models to overfit the data by using too many trainable parameters. We were able to utilize our team's experience using deep neural networks to come up with a robust model that performed a lot closer to the original estimate from stage one and generalized in a predictable manner on unseen data."
"We congratulate our employees who won this difficult competition," said Dr. Algimantas Malickas, owner of Neurotechnology. "These individuals - along with many other excellent employees working on our client projects - demonstrate the qualifications of our Neurotechnology staff and their ability to solve the most complex pattern recognition and neural network training problems."
Neurotechnology is a developer of high-precision algorithms and software based on deep neural network (DNN) and other AI-related technologies. The company offers a range of products for biometric fingerprint, face, iris, palmprint and voice identification as well as AI, computer vision, object recognition and robotics. Drawing from years of academic research in the fields of neuroinformatics, image processing and pattern recognition, Neurotechnology was founded in 1990 in Vilnius, Lithuania and released its first fingerprint identification system in 1991. Since that time the company has released more than 130 products and version upgrades. More than 3000 system integrators, security companies and hardware providers integrate Neurotechnology's algorithms into their products, with millions of customer installations worldwide. Neurotechnology's algorithms also achieved top results in independent technology evaluations including NIST MINEX and IREX.
Jennifer Allen Newton
Bluehouse Consulting Group, Inc.
jennifer (at) bluehousecg (dot) com