Extra neuron and Causal Bayesian networks

  Machine learning

We are working hard to squeeze even more information and inspiration in our meetup. This time two talks! Two speakers, 30 min each!

Does Adding One Neuron Help Real World Networks? - Antonín Hoskovec
The question of presence and effect of local minima in the loss functions commonly used for Machine Learning tasks has always been an interest of mine, which is why when the paper titled "Adding One Neuron Can Eliminate All Bad Local Minima" by Shiyu Liang et al. came out, my senses tingled. It turns out that formally if you take a binary-classification neural network, a dataset and add a single trainable exponential neuron to the output of the network, all the ugly problems with local minima seem to disappear. Here I would like to take you through a bit of why that happens and show you my experiments with the additional neuron, code included.

Causal Bayesian networks - Petr Švarný
Prediction of outcomes or trends can be a very challenging task due to the complexity of relations between the observed variables. A useful tool to create insightful models of such complex systems can be causal Bayesian networks. We look at the basics of causal Bayesian models and play around a little bit with data from the last State of Data Science & Machine Learning report from Kaggle.

About speakers:
Antonín Hoskovec
I am a senior researcher at Rossum, where we are actively working to eliminate manual data entry. My focus is mainly on working with neural networks, I have experience in both designing and deploying Machine Learning models. Other than that I am a Faculty of Nuclear Science and Engineering alumnus who is in the final months of his PhD in Mathematical Engineering.

Petr Švarný
Petr Švarný studies robotics as a PhD student at the Department of Cybernetics at CTU, he also works as a consultant at the Analytics team in Workday, Inc.

Language: English

- 17:45 - 18:00 - Your arrival
- 18:00 - 18:40 - Does Adding One Neuron Help Real World Networks? (and discussion)
- 18:40 - 18:50 - Short break
- 18:50 - 19:30 - Causal Bayesian networks (and discussion)
- 19:30 - 22:00 - Networking in Bitcoin Coffee

Machine Learning Meetups (MLMU) is an independent platform for people interested in Machine Learning, Information Retrieval, Natural Language Processing, Computer Vision, Pattern Recognition, Data Journalism, Artificial Intelligence, Agent Systems and all the related topics. MLMU is a regular community meeting usually consisting of a talk, a discussion and a subsequent networking. At the end of the year 2016, MLMU spread also to Brno and Bratislava. Later on, Košice joined our MLMU family.