Next Meetup

Invited Speaker - Mind and Models: Deep Learning in its Victories and Defeats
Dr. Alan Lockett has been invited as a guest speaker to give a very interesting talk regarding the future of deep learning and how it is reshaping the landscape of data science. Abstract: Under the moniker of deep learning, neural networks have achieved a number of breakthroughs in applied AI over the past five years, from image classification to machine translation. Counter to early narratives suggesting deep learning could only be applied to large datasets, techniques such as representation learning and transfer learning enable successes on small datasets as well. In a certain sense, deep learning automates the construction of ML pipelines, replacing junctures previously constructed by hand with automatically learned interfaces. New APIs and frameworks have improved the accessibility of these techniques, challenging the long-term viability of the traditional data science toolkit composed of SVMs, random forests, decision trees, and logistic regression. Nonetheless, deep learning as presently construed cannot lead to general-purpose AI for very fundamental reasons. A cursory inspection of the outputs of chatbots, machine translation, and language generation reveals that these systems fail to capture or express a consistent narrative thread. Quite simply, deep learning systems lack the qualia of human thought. In their present form, they do not build or maintain simple, consistent models, which makes interpreting or explaining their results difficult. Furthermore, unconstrained and uncurated learning of data replicates statistical biases present in the dataset, which leads to ethical questions regarding their deployment for practical purposes, such as hiring employees or awarding parole. Future work in AI, both industrial and academic, needs to consider how to address these shortcomings by superimposing an artificial mind as a curator over a statistical learning system. In this talk, the successes and challenges of deep learning will be reviewed, followed by discussion of how these challenges might be mitigated and eventually overcome in order to enable general-purpose AI for practical applications. Bio: Alan J. Lockett is the Principal Data Scientist at CS Disco, Inc., a fast-growing legal technology start-up with $60 million invested. He received his Ph.D. in Computer Science from the University of Texas at Austin within the Artificial Intelligence Lab, where he studied neural networks, graphical models, and neuroevolution for applications in games, optimization, and control with Risto Miikkulainen. After his time at UT, he was awarded ad NSF postdoctoral fellowship and worked with Jürgen Schmidhüber at the Dalle Molle Institute for Artificial Intelligence Studies in Lugano, Switzerland on humanoid robotics and deep learning. He is the author of a dozen journal articles and conference papers and holds two patents applying deep learning to legal technology. Agenda: 6:30 Social 7:00 Presentation + QA Location: TBD RSVP: • Seating is limited to the first 140 to RSVP. • Please let any of the ACM Officers know if you have any questions about RSVP.

TBD - Please check with us later

TBD · Austin, TX


    Past Meetups (138)

    What we're about

    Austin Association of Computing Machinery (ACM) Special Interest Group in Knowledge Discovery and Data Mining (SIGKDD). Local Austin chapter of ACM SIGKDD, the premier professional society for machine learning and data mining. This group is specialized to Hadoop based big data machine learning.

    Members (1,551)

    Photos (58)