Education
McGill University and Mila
- Supervised by Dr. Doina Precup and Dr. Simon Gravel
McGill University (Minor Computer Science)
Experience
Google DeepMind, Mountain View California
- Building post-training infrastructure for Gemini models with a focus on distributed reinforcement learning.
- Performance and modelling improvements for various scaling, coding and agentic post-training efforts in Gemini.
- A key member of Gemini 1.0 Pro and Ultra's end-to-end post-training story contributing to with SFT, RL and Eval/Deployment.
- Responsible for the growth, development and promotion of three reports.
DeepMind, Montreal QC
- Building and maintaining centralized reinforcement learning infrastructure used by 100s of researchers across DeepMind.
- Scaling and publishing reinforcement leanring research on affordances(ICML 2020, NeurIPS 2021), and theorem proving (Arxiv 2021.)
- Scaling hierarchical reinforcement learning algorithms on AndroidOS, leading to a novel OCR-free agentic system that could navigate the OS (patent.)
Google Brain, Montreal Canada
- Studied, validated theoretical and published (ICML 2019) policy optimization algorithms and developed novel techniques to visualize why and how they work.
Datalogue, Montreal Canada
- Researched, implemented, and shipped production-level deep conditional random fields for entity recognition, convolutional neural networks for classification, and attention-based recurrent neural networks for machine translation.
- Improved accuracy of main product from 90% to 94% with a 13x reduction in parameters.
Gravel Lab, McGill University
- Used theoretical cancer models to investigate genetic heterogeneity, leading to a publication.
QuantiScience, Montreal
- Engineered an algorithm to extract heart rate variability and infer mental stress from data obtained by the Fitbit Charge HR.
- Launched product to 3 beta testers and demoed in San Francisco as part of the top 10% of the AngelHack HACKcelerator.
Publications
(Full list at Google Scholar)
- Team G., et al.(2023). Gemini: a family of highly capable multimodal models, arXiv preprint arXiv:2312.11805.
- Comanici G., Glaese A., Gergely A., Toyama D., Ahmed Z., Jackson T., Hamel P., Precup D. (2022). Learning how to Interact with a Complex Interface using Hierarchical Reinforcement Learning, arXiv preprint arXiv:2204.10374.
- Khetarpal K., Ahmed Z., Comanici G., Precup D. (2021). Temporally abstract partial models, Advances in Neural Information Processing Systems (NeurIPS).
- Toyama D., Hamel P., Gergely A., Comanici G., Glaese A., Ahmed Z., Jackson T., Mourad S., Precup D. (2021). Androidenv: A reinforcement learning platform for android, arXiv preprint arXiv:2105.13231.
- Firoiu V., Aygun E., Anand A., Ahmed Z., Glorot X., Orseau L., Zhang L., Precup D., Mourad S. (2021). Training a first-order theorem prover from synthetic data, arXiv preprint arXiv:2103.03798.
- Khetarpal K., Ahmed Z., Comanici G., Abel D., Precup D. (2020). What can I do here? A Theory of Affordances in Reinforcement Learning, International Conference on Machine Learning (ICML) 2020.
- Aygün E., Ahmed Z., Anand A., Firoiu V., Glorot X., Orseau L., Precup D., Mourad S. (2020). Learning to prove from synthetic theorems, arXiv preprint arXiv:2006.11259.
- Ahmed Z., Le Roux N., Norouzi M., Schuurmans D. (2019). Understanding the impact of entropy on policy optimization, International Conference on Machine Learning (ICML) 2019.
- Ahmed Z. and Gravel S. (2018). Genetic Diversity in Circulating Tumor Cells, Molecular Biology and Evolution.
- Ahmed Z. (2018). How to Visualize Your Recurrent Neural Network with Attention in Keras, Datalogue Technical Blog.
Awards
- Canada Graduate Scholarship, CIHR | 2017-2018
- Industry Experience Award, NSERC | 2017
- Computational Biology Summer Award, CIHR | 2015 & 2016
- Tomlinson Engagement Award for Mentoring | 2016 & 2017
Selected Projects
- Built data ETL pipelines to ingest 1st-party, high quality city-scale data sources.
- Designed a responsive desktop and mobile app to surface walkable and bikeable city facilities and everday destinations from any location in Toronto and Vancouver.
- Designed and launched a Facebook Messenger bot used by over 25 students to find information about courses and buildings on McGill campus.
- Implemented machine learning classifiers, information retrieval mechanisms, and REGEX-based algorithms resulting in > 80% bot success rate.
- Presented as "Rise of Conversational AI" at Microsoft Chatbot Meetup