HDSI Seminar Series | Spurious or Causal? Studying Failure Modes of Deep Learning and Ways to Fix Them

Title: Spurious or Causal? Studying Failure Modes of Deep Learning and Ways to Fix Them

Abstract: Over the last decade, deep models have enjoyed wide empirical success. However, in practice, these models are not reliable due to their sensitivity against adversarial or natural input distributional shifts as well as a lack of meaningful reasoning behind their predictions. In this talk, I will show that a root cause of these issues is the heavy reliance of deep models on non-causal and spurious features in their inferences. I will then explain our progress in understanding failure modes of deep learning and outline a roadmap towards developing trustworthy learning paradigms.

Bio: Soheil Feizi is an assistant professor in the Computer Science Department at University of Maryland, College Park. Before joining UMD, he was a post-doctoral research scholar at Stanford University. He received his Ph.D. from Massachusetts Institute of Technology (MIT) in EECS with a minor degree in mathematics. He has received an NSF CAREER award in 2020 and is the recipient of several other awards including two best paper awards, a teaching award and multiple faculty awards from industry such as IBM, AWS and Qualcomm. He received a Simons-Berkeley Research Fellowship on deep learning foundations in 2019. He is the recipient of the Ernst Guillemin award for his M.Sc. thesis, as well as the Jacobs Presidential Fellowship and the EECS Great Educators Fellowship at MIT.

HDSI Awards 10 Graduate Prize Fellowships

The Halıcıoğlu Data Science Institute (HDSI) has awarded 10 new Graduate Prize Fellowships to incoming Ph.D. students to support their data-science related research at UC San Diego for the next four years. This influx of data science expertise will boost the research and teaching capabilities for HDSI.

They will work directly with our data science students as teaching assistants for one course, helping strengthen what has become one of the largest data-science education programs in the nation.

The Graduate Prize Fellowships provide substantial financial support to Fellows throughout their four-year award. The awardees are drawn from multiple academic disciplines, in keeping with the multidisciplinary mission of HDSI. Doctoral student appointments will be shared by HDSI and five departments: Bioinformatics and Systems Biology-biomedical informatics track; Computer Science & Engineering (CSE); Mathematics; Cognitive Science; and Electrical and Computer Engineering (ECE).

Awardees are:

  • Fatemeh Amrollahi — Bioinformatics
  • Yeohee Im–ECE
  • Ranak Roy Chowdhury–CSE
  • Matthew Feiglis–Cognitive Science
  • Zhankui He–CSE
  • Side Li–CSE
  • Tara Mirmira–CSE
  • Yanyi Wang–Math
  • Yuyao Wang–Math
  • Weiwei Wu–Math

HDSI Celebrates 1st Year Accomplishments and Vision

Undergraduate scholarship winner presents poster


Capping the celebration of its first year in operations, the Halıcıoğlu Data Science Institute (HDSI) has released a video that spotlights its role as a national leader in data science programs.

The video, which features a day-long symposium held on HDSI’s first anniversary, highlights leading figures in data science research, industry and academia, as well as current students. The focus of the symposium was on both first-year accomplishments, and discussions of the future of the data science field.

In its first year, HDSI demonstrated the historic vision of University of California San Diego by taking the lead on one of the biggest forces in modern life, noted Bob Continetti, UC San Diego Senior Associate Vice Chancellor. With the innovative HDSI approach, he said, “This is how were going to respond by the ever-increasing rate of change around us.”

The symposium was accompanied by another video summing up HDSI first-year accomplishments since launching March 1, 2018. Among the first-year achievements:

  • Establishing itself as one of the largest academic data science programs in the nation
  • Launched a full data science major and minor, drawing students from every class at the university
  • Creating the Institute’s first postdoctoral scholar program
  • Building a core of more than 200 faculty members affiliated across more than 20 academic disciplines, from mathematics to the medical school
  • Attracted more than a dozen industry partners contributing to enhancing educational program, research opportunities and a job pipeline for students

Another leading academician in the data science field appearing on one of the symposium panels was David Culler, UC Berkeley Dean for Data Science and the former Department Chair, Electrical Engineering and Computer Sciences. Culler spoke on the urgent need for responsible data science taught and utilized in research under the guidance of experienced leading universities.

“We’re in a moment where were trying to understand the complicated dynamics of the planet when we also have the ability to see every square yard of it through remote sensing and so forth,” said Culler on the challenges. “What we are seeing are the frontiers of knowledge in essentially every domain are integrated in character.”

The video features undergraduate Luyanda Mdanda accepting his prize for winning the best scientific poster among inaugural group of HDSI Undergraduate Research Scholarship winners.

Mdanda, a junior, was one of more than 40 undergraduates who won competitive scholarships to conduct data science research projects during the academic year. HDSI funded the scholarships that drew students from more than 15 different academic majors, from political science to computer science.