SOCML: Self-Organizing Conference on Machine Learning 2017
I was invited to the Self-Organizing Conference on Machine Learning (SOCML) and attended at the end of November 2017. As per the name, the conference did not have a preset schedule but the participants themselves would put together the agenda on a daily basis and then have sessions. These are just my high level thoughts on the process.
This was my second "Unconference" format. I had attended a South African Treasury Unconference in September. The sessions I attended and wanted to share are next.
ML for Social Good
We had a good number of people (10+) at this session. Few of the people there knew of the Data Science for Social Good program that was started at UChicago and has spread to different universities across the world. When we visited UChicago in October we had a chat on venues to publish in the area (KDD Applied track seemed to be the popular suggestion), but we should be doing more work to get the ML community to connect with the DSSG community. Just within our unit at Modeling and Digital Science at the CSIR, we will have at least 2/6 of our DSIDE student projects using Deep Learning in their solutions.
Notes were taken and shared for this session, I will use them as the basis for the high level topics.
Question: what do you want to get out of the session?
- How do to social good most effectively? (without having to deal with pushback --- produce good & positive social good work and have a broad difference)
- How to get politicians, those in public policy comfortable with machine learning?
- Conversation around empathy and compassionate in context of machine learning
- Role of social sciences vs. data science, multi-disciplinary thought (not discounting those with different backgrounds)
- What problems are more important and suitable for ML? How do we bridge those with problems to solve and those in ML?
- What is the cost of not having a diverse group?
- How to build a better bridge between technical and non-technical people?
- Incentives for attracting top ML talent to social problems? social problems are set to the side because not profitable.
The was an underlying discussion here (In my opinion) on how we can get policy makers to understand what Machine Learning and Artificial Intelligence is. That is, it is not magic. This was an interesting foreshadowing to the other (related) discussion spawned at NIPS2017 on Machine Learning (Deep Learning especially) being Alchemy.
Reinforcement Learning for NLP
Whoa, this was a tough session. We tried, man we tried, to find good connections and interesting questions to discuss but it was tough. I come from an RL background (my PhD) and have over the last 2 years dived into ML + NLP so I wanted to scratch an itch, but it was tough.
Interestingly, Yann Dauphin should have been in this session. He had a poster SOCML on their work on "Deal or No Deal? End-to-End Learning for Negotiation Dialogues" which used RL + NLP to learn how to negotiate between agents (machine or human). I enjoyed the poster.
ML for Social Media
This was a larger crowd and interesting to see the interesting use cases for ML in social media. There are always many. One discussion that spilled out even out of this session was the use of social media data to do user profiling for mobile ads. There are a number of ethical question that rise that the ultimate of this user profiling is for advertisers (or their clients) to identify users who might have addictive personalities. So if the client is a mobile gaming company, with a pay to play model, they want to identify users who will bring them unlimited money. This likely is someone who gets addicted to the game.
ML and Social Science
I have been wanting to move more and more into this space. Especially in trying to work with more social scientists and bring in ML to be able to uncover insights that might have impact with a social science lens.
Increasing Diversity and Inclusiveness
We still have so much to do. The discussion ranged from representation of Women in ML, the challenges and what is going currently in the field to underrepresented minorities. One thing I wanted to highlight from this session was the challenge of treating the minorities who are in ML as though they are representative of the whole community, they likely are in the top 1% of their communities. How do we connect with the rest. This is very true and informed some of our thinking in how we put together the Deep Learning Indaba and doing selection, community building and outreach. I
Black in AI
#BlackInAI It was great to meet the growing community at SOCML and make connections. I was not going to be at NIPS the next week (coming back to do some shepherding for DSIDE), so it was a great opportunity. There is still so much underrepresentation and this is a great initiative by all involved. I will be following up on connecting with more of the community in West, North and East Africa.
This post will likely stay a work in progress as I put in more notes.