#4 Data Science Club, 17/18 Summer
Join us for the forth Data Science Club by Exponea this 17/18 summer semester on April 9th from 4pm to 7pm, hosted by the Faculty of Informatics and Information Technology along with Exponea in Bratislava. The address is Ilkovičova 2, 841 04 Karlova Ves. The session will take place in Aula minor lecture room, room number -1.65. The agenda as well as speaker information can be found below. See you there!
4:00pm: Doors open
4:15pm - 4:20pm: Welcome
4:20pm - 5:20pm: Presentation #1: Building Deep Learning Assistant for Health-Care Specialist
5:30pm - 6:00pm: Presentation #2: Robust Processing of Neural Network Output
6:00pm - 6:45pm: Additional Q&A and Networking
Boris holds a degree in Applied Math from Comenius University in Bratislava. His passion in solving real life problems via Math led him to work as a machine learning specialist at Cognexa. Boris strongly believes in intuition as the most powerful tool, especially in building such blackboxes as neural networks.
Building Deep Learning Assistant for Health-Care Specialist
The working time of a professional doctor, nowadays, is mostly spent in diagnostics - analyzing diagrams, graphs, screens, etc. Using doctors' expertise combined with neural networks results in an efficient software assistant, which decreases time complexity of these routines. Thus, the doctor can focus on other relevant things, such as research or patient care itself.
We will present some intuition regarding the mechanics of neural networks, its benefits and constraints, showing project implementation in tensorflow and high level framework cxflow. All of that will be demonstrated on a case of embryonic growth. We will show you what the best practice in data preparation is and why it is nearly the most important part of machine learning and data analysis. Based on an image timelapse series, we will segment, classify and evaluate actions of the cell, leading to predictions of further progress.
Michal Maly (34) is the director of the AI department in Photoneo. He
worked in the field of reinforcement learning and also spent some
time in computer security. Currently, he spends his efforts giving robots eyes and brains.
Robust Processing of Neural Network Output
Using a neural network can result in correct outputs and some outliers. How to make sure that the robot grasps the object and will
not reach into the void? Robustness of a geometric median helps and is
useful to filter image recognition results, where the candidate's answers are placed in a semantic space.