The Importance of Knowing What We Don’t Know - Bayesianism and Deep Learning
Submitted by Abhijeet Katte (@abhik24) on Saturday, 10 June 2017
Most deep learning models are often viewed as deterministic functions, seen as opaque and different from probabilistic models. But that is not fully true. The probabilistic view of machine learning offers confidence bounds for data analysis and decision making, information that a biologist for example would rely
on to analyse her data, or an autonomous car would use to decide whether to take a turn or brake.
In analysing data or making decisions, it is often necessary to be able to tell
whether a model is certain about its output, being able to ask “maybe I need to use more
diverse data? or change the model? or perhaps be careful when making a decision?”.
This talk will be an introduction to application of Bayesianism in understanding uncertainty in modern deep learning. The key takeaways from the crisp talk will be :
1. Appreciation and understanding of Bayesianism in general
2. Glimpse of how Bayesianism will influence deep learning in the future
3. Application of Bayesianism for better machine learning models
What Do We Mean By Being Bayesian ?
Achievements of Modern Deep Learning
Why Care About Uncertainty
When Is The Probabilistic Approach Essential?
Bayesian Machine Learning
A Few Example and Results
Basic understanding of learning and inference phases of (deep) machine learning. Basic understanding of Bayes’ Rule.
I am Abhijeet Katte. I am a Data Scientist at Hands Free Networks, building intelligent systems for automatically handling support for interconnected and IOT devices. I graduated from Dr. Babasaheb Ambedkar Technological University in 2016 with a degree in computer engineering. Before coming to HFN, I was a part of Data Sciene Team at Locus.sh, a logistics tech company and worked as a research assistant at Indian Institute of Science.