Another interesting event within our Machine Learning series at Morning@Lohika. This time two interesting speakers talked about Physics in ML and ML for Retail.
We started with:
Forecasting of time series using methods of machine learning.
To solve business problems often a Data Scientist is faced with the need to predict a large number of time series. There are many different approaches to solve the problem ranging from smoothing and bringing the series to stationary to predicting by using machine learning methods. During the lecture we will discuss different approaches of forecasting demand for goods in FMCG retail and compare results based on a single set of data.
I work as a Lead Data Scientist at OSA Hybrid Platform, which creates a solution for FMCG industry based on machine learning methods. In particular, I encounter with a wide range of machine learning tasks: classification, clustering, regression, dimensionality reduction. Prior to this, I worked in the financial sector on forecasting proceeds of overdue consumer credit portfolios. I have experience in the fields of studying data more than five years.
And continued with:
Physics-inspired Machine Learning
In recent years, we witnessed the rise of deep neural networks architectures which made an enormous boost in applications as diverse as speech recognition and computer vision. However, in the practical sense, Deep Neural Networks usually are massively redundant, handcrafted and badly optimised systems, build with a large deal of intuition instead of rigorous arguments.
We discussed how to apply physical intuition to some of the Machine Learning problems in order to find better architectures, obtain better performance and universality of the models. Physics has long been linked to machine learning as both fields deal with exponentially large spaces and complex cost-function landscapes. Recently this connection has been revived with new links of deep learning to re-normalization group procedure and quantum many-body wavefunctions. As a result, some powerful physics methods could be applied in Deep Learning to construct and optimize the networks in a controllable way. I will review some examples of how physics manifests itself in machine learning and use-case of a physics-inspired method for a better machine learning.
I am R&D researcher focused on applied Machine Learning, i.e. bio-metric applications, affective computing, signal processing. Hold a PhD in theoretical physics and worked as a researcher in Germany and Israel (Max-Planck Society and Weizmann Institute of Science), developing theory and algorithms for various complex-systems problems spanning from infection propagation in social networks to design of novel quantum states and magnetic materials. My current research interests concern statistical physics of neural networks and application of tensor network algorithms in machine learning.
Leave Your feedback below.