Skip to content

Machine-Learning-Tokyo/__init__

Repository files navigation

Introduction

MLT __init__ is a monthly event led by Jayson Cunanan and J. Miguel Valverde where, similarly to a traditional journal club, a paper is first presented by a volunteer and then discussed among all attendees. Our goal is to give participants good initializations to effectively study and improve their understanding of Deep Learning. We will try to achieve this by:

  • Discussing fundamental papers, whose main ideas are currently implemented on state-of-the-art models.
  • Discussing recent papers.

Sessions

Date Topic Paper Presenter Presentation Video
17/Sep/2022 Inference Optimization LLM.int8() Tim Dettmers Slides Youtube
25/May/2022 CV: GAN SeamlessGAN Carlos Rodríguez-Pardo Slides Youtube
12/Apr/2022 Data Pruning Deep Learning on a Data Diet Karolina Dziugaite Slides Youtube
24/Mar/2022 Bayesian Learning The Bayesian Learning Rule Mohammad Emtiyaz Khan Slides Youtube
26/Jan/2022 NLP: Zero-shot T0 Victor Sanh Slides Youtube
19/Dec/2021 CV: Self-supervised learning SimCLR Kartik Sachdev Slides Youtube
25/Nov/2021 CV: Image Classification ResNet strikes back Ross Wightman Code Youtube
17/Oct/2021 Importance Weighting Rethinking Importance Weighting for Deep Learning under Distribution Shift Nan Lu Slides | Code Youtube
12/Sep/2021 Knowledge Distillation Self-Distillation as Instance-Specific Label Smoothing Mauricio Orbes Slides Youtube
15/Aug/2021 Model Optimization Filter Pruning via Geometric Median J. Miguel Valverde Slides Youtube
18/Jul/2021 CV: Vision Transformers An Image is Worth 16x16 Words Joshua Owoyemi Slides Youtube
13/Jun/2021 NLP: Transformers Attention is all you need Charles Melby-Thompson Keynote Youtube
9/May/2021 NLP: RNN Encoder-Decoder RNN Encoder-Decoder Ana Valeria González Slides Youtube
18/Apr/2021 CV: Object Detection SSD: Single Shot MultiBox Detector Charles Melby-Thompson Slides | Keynote Youtube
14/Mar/2021 CV: Attention in Images Squeeze and Excitation Alisher Abdulkhaev Slides | PwA Youtube
14/Feb/2021 CV: Dilated Convolutions + ASPP DeepLabv2 J. Miguel Valverde Slides | Notebook Youtube
10/Jan/2021 CV: Separable Convolutions Xception Jayson Cunanan Slides | Notebook Youtube

Sessions will be held via Zoom starting at 5pm (JST) / 10am (CET). Check at what time is in your region here.

2021 Summary

MLT __init__ is worldwide! On average we had around 30-50 participants in each of our sessions, joining from, at least, 42 countries. The top 5 countries with the highest number of participants were Japan 🇯🇵, India 🇮🇳, United Kingdom 🇬🇧, Germany 🇩🇪, and Finland 🇫🇮. Thank you for being part of MLT __init__ 🤗 World map 2021

Format

Introduction (5min) + Paper presentation (25min) + Discussion (30min)

We record the introduction and the presentation but not the discussion, allowing participants to interact while protecting their privacy.

For participants

We kindly ask participants to read the paper in advance and to join the session with questions and comments. These questions/comments can be to highlight interesting or unclear parts. For instance: what did you like the most about this paper? What did you learn? What did you not understand?

To make the session more interactive, participants can also ask questions during the presentation. We encourage everyone to use their microphone, but please keep in mind the environmental noise. If you cannot use your microphone or you want to keep your privacy, you are welcome to write in the Zoom chat or Slack channel, and either Jayson or Miguel will read your questions aloud.

For presenters

Presenters will prepare a Powerpoint/Keynote presentation that will be shared in this repository after the session. The presentation should last around 25 mins so that there is enough time for questions and discussion. Inline with the goals of MLT __init__, we encourage presenters to incorporate intuitive visualizations, code, Jupyter notebooks, Colab, and any other material. Finally, please keep in mind that MLT __init__ audience has a very heterogeneous background.Some ideas for the presentation:

  • Background knowledge required to understand the paper.
  • Motivation of the paper, what is the problem that authors try to solve?
  • Contributions of the paper.

Code of Conduct

As this event aims to be interactive, please remember to be kind and respectful to each other. Full code of conduct here.

We want your feedback!

Feedback and contact form: https://forms.gle/jJLWyAMjjVKL8KFRA

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published