ML Conf SF 2018 — Recap

The Machine Learning Conference (MLConf) took place this year at the Hotel Nikko in San Francisco. This was a single-tracked event that gathered experts from the top companies in the machine learning industry to discuss relevant issues in the field and the most up to date techniques to address them.

As people started to arrive, they helped themselves to breakfast and had the opportunity to network with companies that were showcasing their projects in their respective booths.

The morning officially began with the welcome and opening remarks by Courtney Burton, the founder of the MLConf.

The whole event was filled with insightful talks and speakers eager to say that their companies were hiring or happy to talk about any possible collaborations. In this blog post, I will highlight a few of those talks, but if you were not able to attend MLconf this year, you should also check the speaker resource list in this link or all the slideshows here.

The first talk of the day was “Interpretability beyond feature attribution”, where Been Kim from Google Brain talked about Testing with Concept Activation Vectors in order to increase the interpretability of models, and by doing so, using Machine Learning responsibly. This presentation embodied what, by judging the talks of this event, seems to be a general trend among current ML researchers: model interpretability is as important as predictive accuracy.

Later on the day, Franziska Bell from the Natural Language team at Uber talked about two real-world scenarios where NLP algorithms have been applied. The first one was about empowering customer service representatives by suggesting actions and templates based on the intent of the user’s ticket. In this scenario, Bell highlighted Recurrent Decoder Neural Networks that produce more reasonable mistakes, allowing representatives to provide a better service to clients. The second scenario was about generating one-click-chat responses to allow drivers to quickly reply to their riders. In this topic, she discussed unsupervised learning methods applied by Uber that are published in their blog.

Also by Uber, there was Mike Tamir talking about Deep Learning and Fake News. He logically deducted that in order to present a solution to this issue, it had to be converted from an untrackable problem to a trackable problem. One way to do this, instead of verifying if the facts in a journalistic piece are real or not, is to detect whether or not the piece is using emotional manipulation. On a fun note, in the live demo of this application FoxNews got classified as Fake News 😂

Another great technical talk was given by Joan Xiao at Figure Eight. She first presented us with the need for short titles in product recommendation. There are two big approaches to attack this problem: NER and text summarization. Within the text summarization realm, it exists two types of automatic text summarizations: extractive (keeps keywords, remains unchanged) and abstractive (generates a new shorter text). Their experiments showed that there is no significant difference between a short title generated by a human than one generated by an NER algorithm, whereas the text summarization approach performs slightly worse than the other two versions.

One of the most inspiring and futuristic talks of MLconf was “AI for Neuroscience and Neuroscience for AI” by Irina Rish at IBM. This talk underscored the close tie between Neuroscience and Artificial Intelligence, highlighting the fact that both disciplines can deeply enrich each other. She touched upon computational psychiatry and proposed a relevant and futuristic idea about a virtual AI therapist. She concluded by saying that reinforcement learning and deep learning are successful examples where AI learned from neuroscience and that AI still has a lot of room to become more powerful by applying plasticity, neurogenesis, among other concepts from neuroscience.

I would like to end this post by thanking WiMLDS (Reshama Shaikh) for inviting me to MLconf as a blog writer 💕 Also, to the MLconf and the organizers for an amazing event, and for the Twitter book sweepstakes where I happened to be one of the lucky winners.

About me: MS Student in Data Science at the University of San Francisco.




Data Scientist | Mathematician & Journalist

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

Artificial Neural Networks

Bringing Face ID Authentication to the Web

Week 5 — Artificial Pianist

What is Deep Learning: Basics That Every Beginner Should Know

What to Expect From Uncertainty Quantification Methods

SAR to RGB Translation using CycleGAN

AWS EC2 vs EMR vs SageMaker

Cross-gen Spiderman Differtiator

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Viviana Márquez

Viviana Márquez

Data Scientist | Mathematician & Journalist

More from Medium

How To Randomly Sample Data Points (Uniform Distribution)

Why you should consider cluster-robust standard errors in panel and multilevel data analysis

Intuition of the Delta Method in Statistics

How to Stay Up To Date as a Data Scientist