Andrew Ng of DeepLearning.AI announced the winners of the Data-Centric AI competition. The winners in terms of best overall performance are:
Divakar Roy, Shashank Despande, Chris Anderson and Rob Walsh from Innotescus, and Asfandyar Azhar and Nidhish Shah from Synaptic AnN. In the category of the most innovative, the winners were Mohammad Motamedi, Johnson Kuan and the group GoDataDriven | Part of Xebia.
Register for AWS Data & Analytics Conclave>>
Andrew Ng also awarded honorable mentions to Pierre-Louis Bescond and KAIST-AIPRLab (Youngjune Lee, Oh Joon Kwon, Haeju Lee, Joonyoung Kim, Kangwook Lee).
Image source: Andrew Ng | Twitter
- The participants received approximately 3K images of handwritten Roman numerals from 1 to 10. The task was to optimize the performance of the model in the classification of Roman numerals.
- A 52-picture label book to use as a small test set for participants’ own experiments. This book of labels is not used in the final evaluation.
- The model architecture is kept fixed (ResNet50 cutoff) and trained for 100 epochs while the model weights are selected from among the epochs based on the accuracy of the validation set.
- Although the training model and procedure were kept fixed, participants were free to improve the data set and change the training and validation data distributions.
- Adding images was also allowed, but submissions must have less than 10,000 images combined in the training and validation splits.
- Upon submission of the enhanced data set, participants were assessed against a hidden test image set.
- A maximum of five submissions per week (65 in total during the competition) was allowed.
Since only less than 10,000 images were allowed, participants had to focus on getting “good data” in the absence of “big data”. Andrew NG believes that this phenomenon is very common in AI applications for more traditional industries.
Better overall performance
- Divakar Roy, a software engineer working at Findmeashe.com based in Bengaluru.
His interests include software development and testing, problem solving, image and video processing, 3D visualization and measurement, and code porting between C, CUDA, Python and MATLAB.
Roy called the victory the highlight of his professional career. he posted
Image source: Divakar Roy | LinkedIn
- Shashank Despande, Chris Anderson and Rob Walsh from Innotescus (data visualization and image annotation + video platform).
The company said it aims to enable customers to deploy the most reliable and unbiased computer vision models faster by demystifying the preparation and analysis of the most difficult machine learning datasets.
- Asfandyar Azhar and Nidhish Shah of Synaptic AnN.
Shah is a computer science student at Eindhoven University of Technology, and Azhar is taking a combined Bsc / Msc (with distinction) course in Data Science and AI at the same university.
Shah said he has learned a lot about the most innovative approaches to data-centric AI and how to democratize them. He posted on Linkedin:
Image Source: Nidhish Shah | LinkedIn
The most innovative
- Mohammad Motamedi, who works as a Senior Software Engineer – Deep Learning and AI Technology at NVIDIA.
- Johnson Kuan, who is the director of data science and AI / ML at DIRECTV. His role is to lead the implementation of MLOps to accelerate the development / deployment of ML models. It also helps drive the activation and adoption of the latest and most impactful AI / ML techniques.
- GoDataDriven | Part of Xebia – offers data and AI services, consulting and training for the 200 largest companies in the Netherlands and abroad.
What exactly is data-driven AI?
Data-centric AI aims to focus on the quality of the data used to train a model rather than on improving algorithm development. This is the exact opposite of model-centric data, the goal of which is to collect all the data and build a model that is sufficiently capable of handling noise in the data. Andrew Ng argued that this is often much more effective in improving performance.
Myths about data-centric AI
As an emerging field, there is a lot of confusion and myths surrounding data-driven AI. Andrew Ng points out a few:
- Data-centric AI does not solve the critical problem of creating responsible AI
- Just another name for applied machine learning
- Pay more attention to data
- Better data processing
- Only on labeling
- Only works for unstructured data
The top three winners in each of the two categories (Best Overall and Most Innovative Performance) will be invited to a private event with Andrew Ng to share ideas on how to grow the data-centric movement.
Join our Discord server. Be part of an engaging online community. Join here.
Subscribe to our newsletter
Receive the latest updates and relevant offers by sharing your email.