Program at a Glance

THRUSDAY 22 March 
8:00 - 15:00 Registration   8:00 - 15:00 Registration
9:30 - 9:45 Opening Session      
10:00 - 11:00 Keynote Dr. Leo Joscowicz   10:00 - 11:00 Keynote Dr. Reinhard Klette
11:00 - 11:20

Coffee Break

Poster Session I

  11:00 - 11:40

Coffee Break

All-Poster Session

11:20 - 12:20 Keynote Dr. Andrew Barto   11:40 - 12:50 Oral Session III
12:20 - 13:30 Oral Session I      
13:30 - 15:00 Lunch    13:00 - 15:00  Lunch
15:00 - 16:00 Keynote Dr. Jesus Savage   15:00 - 16:00 Keynote Dr. Sajjad Mohsin
16:00 - 16:20

Coffee Break

Poster Session II

  16:00 - 16:15 Coffee Break
16:20 - 17:30 Oral Session II   16:15 - 17:30 Oral Session IV
17:45 - 19:00

City Guided-Tour

  17:30 - 18:00 Closing Session
      19:00 - 20:30 ISICS Dinner


FRIDAY 23 March 
8:30 - 19:00 Guided visit to archaeological site "Chichen Itza"
  (not included in the registration fee)

Keynote Speakers

Wednesday March 21st, 10:00 - 11:00 hrs

Dr. Leo Joscowicz

Title: How is your tumor doing today? Computer-based tumors analysis and follow-up in radiological oncology

Abstract: Radiological follow-up of tumors is the cornerstone of modern oncology. About 25% of the 60 million worldwide yearly CT studies are related to oncology, with a higher proportion for brain MRI studies. Currently, radiologists perform the initial diagnosis and subsequent tumor follow-up manually. This evaluation is tedious, time-consuming, and error-prone, as it varies among radiologists and can be can off by up to 50%. These drawbacks hamper the clinical decision-making process and may lead to sub-optimal or inadequate treatment. In this talk, we will present a new framework for robust, accurate, and automatic or nearly automatic delineation and follow-up of solid tumors in longitudinal multispectral CT and MRI datasets. We will describe new image processing algorithms for brain, lungs, and liver solid tumors and for Plexiform Neurofibromas progression evaluation. We will present the results of our experimental studies and the clinical experience with the software prototype at the Sourasky Medical Center Tel-Aviv.

Wednesday March 21st, 11:20 - 12:20 hrs

Dr. Andrew Barto

Title: Reinforcement Learning: Connections, Surprises, and Challenges

Abstract: Two main reasons that the reinforcement learning (RL) subfield of machine learning has continued to fascinate me for so many years are 1) that it has exposed deep connections between disparate disciplines, ranging from computer science and engineering to psychology and neuroscience, and 2) that it has surprised me in many ways, from unexpected computational efficiency to remarkable parallels with the reward system of animals. In this talk, I will discuss a number of correspondences and surprises that emerged over these many years, including the intimate connection between animal learning and computational methods known as dynamic programming, the striking parallels between popular RL algorithms and the activity of dopamine neurons in the brain, and the remarkable success of programs using RL in surpassing expert human performance in very challenging tasks. Finally, I will comment on RL’s role in the promises and perils of artificial intelligence as it develops over the future.

Wednesday March 21st, 15:00 - 16:00 hrs

Dr. Jesus Savage

Thrusday March 22nd, 10:00 - 11:00 hrs

Dr. Reinhard Klette

Title: Mono-, Bi-, or Tri-nocular Vision for Self-driving Cars

Abstract: Self-driving cars use various sensors, and different camera configurations contribute to the options for sensor configurations. The talk reports about currently obtained results (in joint work with CeRV researchers and PhD students) evaluating different camera configurations for two basic tasks for self-driving cars. Visual odometry derives accurate position data for a moving vehicle based on recorded video data. The use of different camera configurations, and of different analysis strategies (as being possible by the used camera configuration) define particular techniques for visual odometry. A quantitative comparison is provided using KITTI test data. Stixels are a mid-level representation of 3D data in traffic scenes (made famous by vision-based car control systems designed at Daimler A.G.). Stixels help to derive semantic segmentations of traffic scenes, which is a requirement for self-driving cars. Stixels are commonly calculated based on binocular stereo vision. The talk informs about the use of a trinocular approach for increased accuracy, also quantitatively evaluated on KITTI test data. Research at CeRV also applies deep learning for solving various object detection, tracking, or scene analysis tasks. However, the use of mathematically defined models appears to be the adequate way for visual odometry and stixel calculation.

Thrusday March 22nd, 15:00 - 16:00 hrs

Dr. Sajjad Mohsin