|
|
[[Home](Home)]
|
|
|
|
|
|
---
|
|
|
|
|
|
### Talk by [Emre Neftci](http://nmi-lab.org/people-apphook/eneftci/): "Data and Power Efficient Intelligence with Neuromorphic Hardware"
|
|
|
* When: Tuesday, 27 August 2019, 1:00pm
|
|
|
* Where: Jülich Supercomputing Centre, Rotunda, building 16.4, room 301
|
|
|
|
|
|
The potential of machine learning and deep learning to advance artificial intelligence is driving a quest to build dedicated systems that accelerate such workloads at a large scale and in an autonomous fashion. A natural approach is to take inspiration from the brain by building neuromorphic hardware that emulates the biological processes of the brain using digital or mixed-signal technologies.<br>
|
|
|
In this talk, I will present interdisciplinary approaches anchored in machine learning theory and computational neurosciences that enable the applicability of neuromorphic technologies to real-world, human-centric tasks. In particular I’ll discuss the following related challenges and their possible solutions:<br>
|
|
|
(1) The models and tools of deep learning apply to neuromorphic hardware, but physical implementations of neural networks call for novel, continual and local learning algorithms;<br>
|
|
|
(2) Neuromorphic technologies have potential advantages over conventional computers on tasks where real-time adaptability, autonomy or energy efficiency are necessary, but applications and benchmarks benefiting from these qualities are not yet identified;<br>
|
|
|
(3) Challenges in memory technologies, compounded by a tradition of bottom-up approaches in the field and the lack of large-scale simulation environments block the road to major breakthroughs.<br>
|
|
|
The recent algorithmic results I will present solve some of these challenges and pave the way toward the co-design of brain-inspired computing systems and algorithms with a mathematical viewpoint. These solutions enable the roadmap towards building a software framework for neuromorphic hardware with a Tensorflow-like workflow and leveraging the scalable, distributed, low-latency and energy- efficient nature of neuromorphic hardware.
|
|
|
|
|
|
### Talk by [Frank Rudzicz](http://www.cs.toronto.edu/~frank/): "Explainable AI — Interpretability, brains, policies and politics"
|
|
|
* When: 25th July 2019, at 11:00 am
|
|
|
* Where: INM seminar room, building 15.9, room 4001b
|
|
|
|
|
|
In this talk, we take a path through several different approaches to explainability in machine learning. First, we talk about categories of explainability, then we discuss approaches to relevance ranking in terms of engineered features and in terms of heat maps in images through deep Taylor expansion. We then provide a use case of a recent publication on using machine learning with MEG data, and suggest that explainability in brain data has room for improvement. Time permitting, we will briefly cover how explainable AI may help to overcome regulatory and cultural issues in healthcare and therefore accelerate the use of AI methods in practice.
|
|
|
|
|
|
### Talk by [Mateusz Kozinski ](https://cvlab.epfl.ch/): "Learning to segment 3D linear structures with 2D annotations"
|
|
|
* When: 18th July 2019, at 1:30 pm
|
|
|
* Where: INM seminar room, building 15.9, room 4001b
|
|
|
|
|
|
We propose a loss function for training a Deep Neural Network (DNN) to segment volumetric data, that accommodates ground truth annotations of 2D projections of the training volumes, instead of annotations of the 3D volumes themselves. In consequence, we significantly decrease the amount of annotations needed for a given training set. We apply the proposed loss to train DNNs for segmentation of vascular and neural networks in microscopy images and demonstrate only a marginal accuracy loss associated to the significant reduction of the annotation effort. The lower labor cost of deploying DNNs, brought in by our method, can contribute to a wide adoption of these techniques for analysis of 3D images of linear structures.
|
|
|
|
|
|
### Lab Visit at JSC - SimLab Neuroscience
|
|
|
* When: Thursday, **9 May 2019, 2 pm**
|
|
|
* Where: JSC, Rotunde, building 16.4, room 301
|
... | ... | |