MLT Talks

Slides, videos and other resources from MLT Talks.

• An introduction to transfer learning in NLP and HuggingFace, Thomas Wolf, Co-Founder and CSO at HuggingFace (October 2020)
• Design and Testing patterns for Machine Learning, Chris Gerpheide, CTO at Bespoke Inc. (September 2020)
• Open-source for quantum technologies, Nathan Shammah, RIKEN (February 2020)
• NLP Business Applications, Cedric Wagrez, Gengo, Lionbridge AI (February 2020)
• 'Scikit-project': How open source is empowering open science – and vice versa, Nathan Shammah, Postdoctoral Researcher, Theoretical Quantum Physics Laboratory, RIKEN (November 2019)
• Python tools for Deep Learning--Chainer, CuPy, and Optuna, Crissman Loomis, Engineer / Business Development, Preferred Networks, Inc. (September 2019)
• Creating a Chatbot from Scratch, Chris Gerpheide, Bespoke Inc. (September 2019)
• Augmented Reality and Artificial Intelligence, Mark Billinghurst, School of Information Technology and Mathematical Sciences, University of South Australia (September 2019)
• Decision (Neuro)Science in Society, Rei Akaishi, RIKEN Center for Brain Science – Toyota Collaboration Center (September 2019)
• The many machine learning ways of Google Translate, Keith Stevens, Google Japan (August 2019)
• Artificial Life, Lana Sinapayen, Sony CSL, ELSI (August 2019)
• Brain-Machine-Interfaces, Talk & Discussion, Antonio Lozano, UPCT, Spain (July 2019)
• Getting started with TensorFlow 2.0, Josh Gordon, Developer Advocate at Google AI (April 2019)
• Current capabilties, limitations and future directions of deep learning, François Chollet, Creator of Keras (March 2019)
• Deep Neural Networks for Video Applications, Alex Conway, NumberBoost (March 2019)
• Beyond Supervised Driving Adrien Gaidon, Toyota Research Institute (March 2019)

Thomas Wolf (HuggingFace)

An introduction to transfer learning in NLP and HuggingFace

Talk

MLT welcomed Thomas Wolf, Co-founder and Chief Science Officer at HuggingFace, for a talk on "An introduction to transfer learning in NLP and HuggingFace". Thomas introduces the recent breakthroughs in NLP that resulted from the combination of Transfer Learning schemes and Transformer architectures. The second part of the talk is dedicated to an introduction of the open-source tools released by HuggingFace, in particular Transformers, Tokenizers and Datasets libraries and models. Read more: https://huggingface.co/

Speaker Bio

Thomas Wolf is co-founder and Chief Science Officer of HuggingFace. His team is on a mission to catalyze and democratize NLP research. Prior to HuggingFace, Thomas gained a Ph.D. in physics, and later a law degree. He worked as a physics researcher and a European Patent Attorney. https://thomwolf.io/

• Slides
• Talk (YouTube)

Chris Gerpheide (Bespoke)

Design and Testing patterns for Machine Learning

Talk

FactoryFactories and so can you – OO Design and Testing patterns for Machine Learning Object-oriented programming has long been known for its design patterns, which help coders write maintainable, high-quality code. In this talk, Chris talks about a number of design patterns that are particularly useful for machine learning applications, including why and how you can implement them yourself. She also covers strategies for writing unit and integration tests for the machine learning code.

Speaker Bio

Chris Gerpheide is the CTO at Bespoke Inc. https://www.be-spoke.io. Before joining Bespoke, she was an engineering manager at Amazon Web Services in the Netherlands. In her free time, she enjoys learning Japanese, hiking, and playing the violin.

• Slides
• Code

Nathan Shammah (RIKEN)

Open-source for quantum technologies

Talk

In past decades there has been considerable advancement both in theory and in experiments in research laboratories around the world. More recently, there has been an increased effort toward spinning out technology from our scientific understanding of quantum theory. The software stack interfacing with these early prototypes of quantum machines is heavily relying on the open-source paradigm. Nathan will review software and hardware trends, and give an intro to quantum information processing. He’ll speak about the quantum toolbox in Python (QuTiP), general community efforts aimed at creating a sustainable ecosystem for open-source projects and provide some examples of how QC and ML are beginning to interact with each other.

Speaker Bio

Nathan Shammah is a Postdoctoral Research Scientist at RIKEN, Japan’s national lab, in the Quantum Physics Theory Lab, where he investigates the dynamics of open quantum systems. Lead developer of QuTiP, the quantum toolbox in Python, he also writes the quantum tech newsletter.

Cedric Wagrez, Gengo, Lionbridge AI

NLP Business Applications

Talk

In this talk, Cedric provided an overview of NLP technology, and explained how it is being used across a wide range of services. He showed case studies, and explained the challenges ahead, both globally and in Japan.

Speaker Bio

Cedric has been living in Japan for over 15 years, and was appointed VP of Operations at Gengo (http://gengo.com) 2 years ago. Following Gengo's acquisition by Lionbridge, he recently took over another challenge: developing Lionbridge's AI services offering in Japan, centered around high-quality data for AI applications (https://lionbridge.ai).

Nathan Shammah (RIKEN)

'Scikit-project': How open source is empowering open science – and vice versa

Talk

Nathan shows how open-source pipelines are accelerating scientific discovery, by empowering not only reproducibility of research results but also generalizability of methods. He addresses the rise of open source in scientific research in quantum physics and quantum information and introduce 'scikit-project' a cookbook with best practices for (data) scientists.

Speaker Bio

Nathan Shammah is a Postdoctoral Research Scientist at RIKEN, Japan’s national lab, in the Quantum Physics Theory Lab, where he investigates the dynamics of open quantum systems. Lead developer of QuTiP, the quantum toolbox in Python, he also writes the quantum tech newsletter, https://medium.com/quantum-tech.

Crissman Loomis (Preferred Networks, Inc.)

Python tools for Deep Learning--Chainer, CuPy, and Optuna

Talk

Starting with an introduction to the Deep Learning framework Chainer, Crissman will present how to structure basic Deep Learning models. Also covered will be CuPy, a NumPy-like API for calculations on NVIDIA GPUS, and Optuna, a hyper parameter optimization library. Attendees will be able to do hands-on work using Google Colab and to understand the uses and benefits of all three open source projects. https://preferred.jp/en/

Speaker Bio

Crissman has worked at Preferred Networks on the Chainer team for over two years, focusing on improving the documentation for Chainer and giving presentations on Chainer at Open Data Science Conferences, Euro SciPy, PyCon, GTC, and other venues. His ODSC West workshop on Chainer was selected as one of the top 10 workshops for learning Machine Learning.

Chris Gerpheide (Bespoke Inc.)

Creating a Chatbot from Scratch

Talk

Ever wondered how chatbots work? Chris Gerpheide is leading the engineering team at Bespoke, the creators of Bebot, a chatbot for hospitality and tourism in Japan. Bespoke develops all of their technology in house; during the workshop, Chris will show you some simple machine learning and natural language processing techniques to create the guts of a basic chatbot from scratch. This is an interactive workshop, where each participant can create their own simple chatbot backend in Python using their own laptops.

Speaker Bio

Chris Gerpheide leads the engineering team at Bespoke Inc. Before joining Bespoke, she was engineering manager at Amazon Web Services. In her free time, she enjoys learning Japanese, hiking, and teaching children programming.

Mark Billinghurst (University of South Australia)

Augmented Reality and Artificial Intelligence

Rei Akaishi (RIKEN CBS)

Decision (Neuro)Science in Society

Talk

Data and its analysis create value by assisting humans in the decision making process. The mergence of decision science and data science opens new perspectives and opportunities for research and industry. In this talk Rei Akaishi gives an introduction to decision neuroscience and behavioral economics and discusses the synthesis of decision science and data science with the audience.

Speaker Bio

Rei Akaishi obtained his PhD from Graduate School of Medicine, the University of Tokyo. He did his post-doctoral jobs in University of Oxford, Tokyo Metropolitan Institute of Medical Science, University of Rochester, Center for Information and Neural Networks and conducted research projects on decision making of humans and animals. He is currently a Unit Leader in RIKEN Center for Brain Science.

Keith Stevens (Google Japan)

The many machine learning ways of Google Translate

Talk

Keith covers recent innovations in machine translation published by the team behind Google Translate. Learn how these models are getting bigger, how they're solving more complex tasks like simultaneous translation, and how embedding methods can be used to find and clean data. He goes into depth about what problem each of these efforts are trying to solve, hear what new approaches are being used, and detail lessons learned while working on these challenges.

Speaker Bio

Keith Stevens has been on Google Translate for nearly years 7 years. His primary focus has been finding or creating the best data possible for improving Google Translate. This work has ranged from Crowdsourcing systems to neural based data cleaning techniques and automated parallel data mining.

Lana Sinapayen (Sony CSL, ELSI)

Artificial Life

Talk

Artificial Life is a little known field spanning disciplines as diverse as Informatics, Chemistry and Robotics. Lana will give a brief overview of the field and its main research questions.

Speaker Bio

LANA SINAPAYEN is and Artificial Life and Artificial Intelligence researcher. Her main interests are the emergence of cognitive functions such as predictive coding, and evolutionary dynamics leading to open ended systems. Artificial Life is a little known field spanning disciplines as diverse as Informatics, Chemistry and Robotics. In her talk she will give a brief overview of the field and its main research questions.

Antonio Lozano (UPCT Spain)

Brain-Machine-Interfaces

Talk

Neural engineering and AI are making rapid progress, and new opportunities for synergy arise. We are facing the challenge of creating neuroprosthesis designed to interface with the visual cortex in order to restore a limited but useful visual sense to these blind patients. This talk will look at neural engineering, some of the advances that highly multidisciplinary teams of scientists and engineers are making in different parts of the world, including Neuralink's new brain interfaces, and some of the future challenges that we face in order to improve people's lives. Antonio will also introduce a new framework -NeuroLight- based on Convolutional Neural Networks, created to encode visual information and transmit this it in a meaningful way to the human brain through a neural prosthesis.

Speaker Bio

ANTONIO LOZANO is an Industrial Engineer, and Intel's Software Innovator. He's pursuing a PhD in Information and Communications Engineering at UPCT, Spain, collaborating with the Neuroengineering Biomedical Research Group at UMH, towards the goal of creating a cortical visual neuroprosthesis for the blind. Recently, he visited Tokyo Institute of Technology as a visiting junior fellow researcher.

• Development of a Cortical Visual Neuroprosthesis for the Blind (CORTIVIS project)
• Biomedical Neuroengineering Research Group, UMH, Spain

Josh Gordon (Google AI)

Getting started with TensorFlow 2.0

Talk

TensorFlow 2.0 is all about ease of use. In this 45 minute talk, I'll cover best practices for beginners and experts, and point you to the latest code examples you can try for each. I'll cover the Keras Sequential, Functional, and Subclassing APIs, as well as built-in training loops, and how to write a custom training loop using a GradientTape. To wrap it up, I'll give a quick summary of a few announcements and updates from the TensorFlow Developer Summit.

Speaker Bio

Josh Gordon is a Developer Advocate at Google AI, and also teaches Applied Deep Learning at Columbia University, and Machine Learning at Pace University. He has over a decade of machine learning experience to share. You can find him on Twitter at @random_forests.

François Chollet (Creator of Keras, Google)

Current capabilties, limitations and future directions of deep learning

Talk

Deep learning has had amazing successes in recent years. But can it lead to Strong AI? The goal of this talk is to zoom out a little bit -- to look at how deep learning really works, to look at its current limits, and to try to see the road ahead for AI.

Speaker Bio

François Chollet is the Creator of Keras (keras.io), a leading deep learning API, and author of the textbook "Deep Learning with Python". He is also a machine learning researcher at Google Brain and a contributor to the TensorFlow machine learning platform.

Alex Conway (NumberBoost)

Deep Neural Networks for Video Applications

Talk

Most CCTV video cameras exist as a sort of time machine for insurance purposes. Deep neural networks make it easy to convert video into data which can then be used to trigger real-time anomaly alerts and optimize complex business processes. Deep learning can also be used in academic research to speed up labeling of video recorded from the point of view of animals wearing go-pros. This talk will present some theory of deep neural networks for video applications as well as academic research and several applied real-world industrial examples.

Speaker Bio

Alex is the Founder and Head of Data Science at NumberBoost, a startup based in Cape Town that builds custom A.I. solutions focused on real-time computer vision using deep learning, edge computing and privacy-preserving federated machine learning. NumberBoost has won startup competitions with MultiChoice, Mercedes-Benz, Lloyd's Register in London and NTT Data Japan. He organizes the Cape Town Machine Learning and Deep Learning Meetup groups and chaired the organizing committee for the 2018 Deep Learning IndabaX conference in Cape Town.

Adrien Gaidon (TRI)

Beyond Supervised Driving

Talk

Crowd-sourced steering does not sound as appealing as automated driving. We need to go beyond supervised learning for automated driving, including for computer vision problems seeing great progress with strong supervision today. First, we will motivate why this is required for long-term large-scale autonomous robots. Second, we will discuss recent state-of-the-art results obtained in the ML team at Toyota Research Institute (TRI) for unsupervised domain adaptation from simulation and self-supervised depth and pose prediction from monocular imagery. Finally, I will talk about how we actually scale to large datasets using our cloud infrastructure and distributed deep learning.

Speaker Bio

Adrien Gaidon is the Manager of the Machine Learning team and a Senior Research Scientist at the Toyota Research Institute (TRI) in Los Altos, CA, USA, working on open problems in world-scale learning for autonomous driving. He received his PhD from Microsoft Research - Inria Paris in 2012 and has over a decade of experience in academic and industrial Computer Vision, with over 30 publications, top entries in international Computer Vision competitions, multiple best reviewer awards, international press coverage for his work on Deep Learning with simulation, and was a guest editor for the International Journal of Computer Vision. You can find him on LinkedIn (https://www.linkedin.com/in/adrien-gaidon-63ab2358/) and Twitter (www.twitter.com/adnothing).

Recommendation in Production

Talk

We will discuss about an entire life cycle of the deployment of Recommendation Systems in production environments. From data to algorithms: We will look at building data pipelines and architectures that will allow us to experiment and deploy in production in an easy and fast way. (https://github.com/Machine-Learning-Tokyo/MLT_Talks/blob/master/slides/Starfish.pdf)

Speaker Bio

Shripad Deshmukh is Senior Engineer in Data Science Team at U-Next japan working on recommendaiton engine and production system of recommendaiton system. You can find hime on LinkedIN (https://www.linkedin.com/in/shripad-deshmukh-49062016/).