Stony Brook University’s Institute for AI-Driven Discovery and Innovation has been working tirelessly to study and present new research to the world. As a result, this year the Institute had seven papers accepted at the most distinguished conference for machine learning in the world: The Conference and Workshop on Neural Information Processing Systems (NeurIPS).
Stony Brook researchers of these accepted papers are:
Department of Computer Science
Department of Biomedical Informatics
Renaissance School of Medicine, College of Arts and Sciences, Department of Neurobiology and Behavior
As a global cornerstone of the artificial intelligence community, NeurIPS annually selects the most exceptional AI research papers. This year the Conference received almost 10,000 full paper submissions, with an acceptance rate of only 20.1 percent. NeurIPS has become increasingly more competitive. To have seven papers selected puts Stony Brook among the top universities internationally for AI research.
“NeurIPS is the premier conference in machine learning, with prominence throughout science and engineering,” said AI Institute Director Professor Steven Skiena. “Just to be able to attend the conference, let alone have seven papers be accepted is exceptional, and testament to the relevance and impact of the research underway here at Stony Brook.”
Il Memming Park, associate professor on Institute faculty, had two papers accepted this year at NeurIPS. His research on rescuing neural spike train models is a transformative contribution to the artificial intelligence and neuroscience communities. Park and his team utilized skills from machine learning and neuroscience to create a method that tracks stimuli and neural responses.
“’Rescuing Neural Spike Train Models from Bad MLE’ addresses an important problem of evaluating neural spike train models,” said Park. “In collaboration with Diego Arribas, a visiting scholar from Argentina, and Dr. Yuan Zhao, we’ve been able to create a new evaluation metric and demonstrated its usefulness.”
In essence, Professor Park and his collaborators have created a model that predicts more likely outcomes for these very responses. Their research produces data that had been previously ignored in other models, and therefore their new model proves to be more reliable. The model is so impressive that its applications do not stop at neuroscience; it has the capacity to reach areas like the social sciences, geophysics, astrophysics and finance.
Minh Haoi Nguyen, associate professor in the Department of Computer Science, also had two papers accepted. “Detecting Hands and Recognizing Physical Contact in the Wild” tackles the issue of ambiguity in hand detection. Because of the variety when it comes to hand shapes, detecting and recognizing hands, especially when their image is in motion or slightly obstructed, is very difficult. Nguyen and his collaborators successfully created a model that produced positive and accurate feedback for hand detection.
The team even took this a step further, going on to implement predictive methods for hand orientation. “We hope our work will spark the community’s interest in addressing this important problem, which has many potential applications, including harassment detection and contamination prevention,”said Nguyen. The success of this research is a new and useful contribution to the Artificial Intelligence and societal communities.
The Conference will take place in mid-December. While its presentation will take place solely in a virtual format, the recognition is equally as important. “Seven papers at NeurIPS 2020 places our AI research among the top schools in the world,” said Professor Skiena.
Following is a list of all seven accepted papers:
“Deep Variational Instance Segmentation” by Jialin Yuan (Oregon State University) · Chao Chen (Stony Brook University) · Fuxin Li (Oregon State University)
“On 1/n Neural Representation and Robustness” by Josue Nassar (Stony Brook University) · Piotr Sokol (Stony Brook University) · Sueyeon Chung (Columbia University) · Kenneth D Harris (UCL) · Il Memming Park (Stony Brook University)
“A Topological Filter for Learning with Label Noise” by Pengxiang Wu (Rutgers University) · Songzhu Zheng (Stony Brook University) · Mayank Goswami (Queens College of CUNY) · Dimitris Metaxas (Rutgers University) · Chao Chen (Stony Brook University)
“Detecting Hands and Recognizing Physical Contact in the Wild” by Supreeth Narasimhaswamy (Stony Brook University) · Trung Nguyen (VinAI) · Minh Hoai Nguyen (Stony Brook University)
“Distribution Matching for Crowd Counting” by Boyu Wang (Stony Brook University) · Huidong Liu (Stony Brook University) · Dimitris Samaras (Stony Brook University) · Minh Hoai Nguyen (Stony Brook University)
“AViD Dataset: Anonymized Videos from Diverse Countries” by Anthony JPiergiovanni (Indiana University) · Michael S. Ryoo (Stony Brook University)
“Rescuing Neural Spike Train Models from Bad MLE” by Diego Arribas (Stony Brook University) · Yuan Zhao (Stony Brook University) · Il Memming Park (Stony Brook University)
— Alyssa Dey, Communications Assistant