Notes from the Floor: Women in Machine Learning Workshop 2019

This year’s Women in Machine Learning (WiML) workshop took place on Monday (December 9) in Vancouver, British Columbia, and featured key industry heavyweights at the cutting edge of artificial intelligence (AI).

With speakers from Google’s (NASDAQ:GOOG) Google AI, the University of California Berkeley and Uber (NYSE:UBER), the one day conference was also rumored to have deep-learning luminary Yann Lecun, who won the Turing Award last year, in attendance.

In its 14th year, the conference has come a long way. With under 100 attendees last year in Montreal, Quebec, this year’s Vancouver edition brought a tenfold increase to over 1,000.

“Security will be one of the biggest challenges in deploying AI,” says Dawn Song, expert in security and AI at the University of California Berkeley during the Women in Machine Learning Conference in Vancouver, BC. Stay tuned for more coverage throughout the day #wiml2019 pic.twitter.com/ozLpEI97DN

— Tech Investing News (@INN_Technology) December 9, 2019

During the first talk of the day, Dawn Song, a professor at the University of California Berkeley, spoke about the intersection of security and AI.

“Security will be one of the biggest challenges in deploying AI,” said Song. She went on to discuss how there is no sufficient defense against attackers present in machine learning. “Attackers can extract security numbers from the language model, or training data in AI,” she added.

Song is also the CEO of Oasis Labs, a startup that is creating cloud-based security systems with blockchain technology. Deeplearning.ai has called her one of the heroes of deep learning.

Snapshots from the floor #WiML2019. Speakers from @Uber @salesforce @GoogleAI and several others discuss selection bias and reinforcement learning in AI. pic.twitter.com/lwUqkysAvF

— Tech Investing News (@INN_Technology) December 9, 2019

Following her talk, speakers included Xinyi Chen, a doctoral student at Princeton University working with Google AI, and Xanda Schofield, an assistant professor at Harvey Mudd College.

In a talk titled “When Text Mining Isn’t a Piece of Cake,” Schofield spoke about her collaborative work with data journalists, sociologists and economists.

Xanda Schofield discusses how AI text mining is used for @Spotify playlists. She later discusses how user interfaces matter to leverage usability in AI. pic.twitter.com/eLAAux2J89

— Tech Investing News (@INN_Technology) December 9, 2019

Schofield went on to discuss how the text-mining process typically unfolds. To analyze a broad set of text, she applies classification algorithms to text to get labeled information records. Breaking down her process, she described it like this: find a text collection, process the text, learn a model and interpret the model.

For example, text mining has been applied to Spotify (NYSE:SPOT) playlists. Applying three types of machine learning — raw audio, collaborative filtering and natural language processing — it creates curated playlists for millions of its users.

#WiML2019 partner conference NeurIPS features sponsors from @DeepMindAI @Sony and Bloomberg @technology pic.twitter.com/DXmOXAQwWL

— Tech Investing News (@INN_Technology) December 9, 2019

Coinciding with the WiML workshop was NeurIPS, an international conference on neural information processing systems. NeurIPS features some of the most advanced research in machine learning in the field today. Google, OpenAI and Facebook (NASDAQ:FB) Artificial Intelligence were present, in addition to several AI investment pioneers such as DE Shaw and Two Sigma.

In a talk surrounding bias in AI, PhD student at New York University (NYU) Margarita Boyarskaya discussed how difficult it is to achieve fairness in machine learning. “Most methods rely on untestable assumptions,” she said, posing it as an opportunity instead of an ominous likelihood in the future.

“There’s an empathy gap between the technology being created and society that’s being impacted,” Kathy Baxter, architect of ethical AI practice at @salesforce $CRM, says in a talk at #WiML2019

— Tech Investing News (@INN_Technology) December 9, 2019

Kathy Baxter @baxterkb of @salesforce gives an invited talk on ‘Creating the world we want: How to build an ethical AI practice’ #WiML2019 pic.twitter.com/h1C7pfecJG

— WiML (@WiMLworkshop) December 9, 2019

While Boyarskaya discussed bias, a following talk covered ethics in AI. After talks from academics at NYU and Carnegie Mellon, Kathy Baxter, architect of ethical AI practice at Salesforce (NASDAQ:CRM), talked about the role AI plays in company values and ethical frameworks.

“There’s an empathy gap between the technology that is being created and society that’s being impacted,” she said. She described how Salesforce has machine -earning courses and classes for employees, while impressing how AI needs to safeguard human rights. She described how company data belongs to the company itself, rather than Salesforce.

Later in the day, speakers from MIT Media Labs and UberAI spoke about deep learning in image recognition and the concept of attention as explanation. Researchers from the University of Pittsburgh described how attention as explanation can be used to deter from bias, through evaluating or auditing machine-learning models to check for fairness, bias and accountability.

Looking forward, developing the right models and user interfaces will be central to AI development to allow individuals to work with greater ease and agency in AI systems.

Don’t forget to follow us @INN_Technology for real-time news updates!

Securities Disclosure: I, Dorothy Neufeld, hold no direct investment interest in any company mentioned in this article.