Room 416-419
Seminar on Human & Machine Intelligence || We are virtual! See details below.
ZOOM LINK TO THIS VIRTUAL MEETING: https://us02web.zoom.us/j/86386159843
Monday, May 18 (*** 17:00-19:00 *** NOTE TIME CHANGE)
Organized by:
Prof. Dr. Matthias Kaschube (FIAS & Goethe University Frankfurt)
Prof. Dr. Kristian Kersting (TU Darmstadt)
Prof. Dr. Stefan Kramer (Johannes Gutenberg University of Mainz)
Prof. Dr. Visvanathan Ramesh (Goethe University Frankfurt)
Prof. Dr. Gemma Roig (Goethe University Frankfurt)
Prof. Dr. Constantin A. Rothkopf (TU Darmstadt & FIAS)
Prof. Dr. Jochen Triesch (FIAS & Goethe University Frankfurt)
The disciplines of machine learning, brain sciences and complex systems engineering for AI is constantly evolving with new state-of-the-art techniques being introduced on a regular basis. There is an emerging ecosystem in the Frankfurt Rhein-Main region focusing on transdisciplinary aspects of intelligent systems where various facets of the development and impact of AI on sciences, humanities, and vice versa are being studied. Our biweekly seminar complements other seminars in the region and is dedicated to fast-paced developments in the field and is particularly aimed at graduate students, postdocs, machine-learning researchers and cognitive scientists. The initial focus of the seminar will be on the latest developments in modern deep-learning applications, focusing mainly on domain-specific applications such as NLP, vision, and audio. We also plan to explore theoretical aspects of interest to development of third-wave AI systems that are context sensitive, enable interpretability/explainability of how results are arrived at, and provide transparency thus enabling fundamental insights on limits of algorithm designs in context. Other topics include the intersection between computational neuroscience, mathematical psychology and machine learning. Every meeting a presenter will provide a short introduction to the topic, followed by a description of the main findings of one or more papers which define the state-of-the art in the field.
Click here to join our mailing list
Click for the yearly calendar for the seminar
Click here to view/subscribe to our topic list
Next *VIRTUAL* meeting:
ZOOM LINK TO THIS VIRTUAL MEETING: https://us02web.zoom.us/j/86386159843
Mondays, May 18 (*** 17:00-19:00 ***) NOTE THE TIME CHANGE FOR THIS WEEK!!!
Title: Task-specific Vision DNN Models and Their Relation for Explaining Different Areas of the Visual Cortex
Prof. Dr. Gemma Roig (Goethe University Frankfurt)
Abstract: Deep Neural Networks (DNNs) are state-of-the-art models for many vision tasks. We propose an approach to assess the relationship between visual tasks and their task-specific models. Our method uses Representation Similarity Analysis (RSA), which is commonly used to find a correlation between neuronal responses from brain data and models. With RSA we obtain a similarity score among tasks by computing correlations between models trained on different tasks. We demonstrate the effectiveness and efficiency of our method to generating task taxonomy on the Taskonomy dataset. Also, results on PASCAL VOC suggest that initializing the models trained on tasks with higher similarity score show higher transfer learning performance. Finally, we explore the power of DNNs trained on 2D, 3D, and semantic visual tasks as a tool to gain insights into functions of visual brain areas (early visual cortex (EVC), OPA and PPA). We find that EVC representation is more similar to early layers of all DNNs and deeper layers of 2D-task DNNs. OPA representation is more similar to deeper layers of 3D DNNs, whereas PPA representation to deeper layers of semantic DNNs. We extend our study to performing searchlight analysis using such task specific DNN representations to generate task-specificity maps of the visual cortex. Our findings suggest that DNNs trained on a diverse set of visual tasks can be used to gain insights into functions of the visual cortex. This method has the potential to be applied beyond visual brain areas.
We will discuss the following papers, which you are welcome to read before the meeting:
- Representation similarity analysis between DNNs (application to transfer learning).
- DNNs for explaining functionality of different brain regions.
- Algonauts project I: https://ccneuro.org/2019/proceedings/0000264.pd
- Algonauts project II: https://www.nature.com/articles/s42256-019-0127-z
- Unravelling Representations in Scene-selective Brain Regions Using Scene Parsing Deep Neural Networks
Previous meetings:
May 4, 2020: Dr. Steffen Eger (TU Darmstadt) on "On the State-of-the-Art in Evaluation for Machine Translation and Summarization"
March 2, 2020: Dr. Peter Harrison, Federico Adolfi, Raja Marjieh and Dr. Nori Jacoby (MPIEA) on "Deep generative models in audition and music: Towards a unified framework"
Februrary 17, 2020: Prof. Dr. Alexander Gepperth (University of Applied Sciences Fulda) on "Real-world continuous learning"
Februrary 3, 2020: Timm Hess (Goethe University) on "Evaluation of Continual Machine Learning using Systematic Simulation"
January 13, 2020: Prof. Dr. Jochen Triesch (FIAS & Goethe University Frankfurt) on "Active Efficient Coding"
December 2, 2019: Prof. Dr. Christoph von der Malsburg (FIAS) on "AI: How to get out of the Human Shadow?"