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

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__ 🤗

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