
Slides: Distributed Training for ML
Explore distributed training techniques through this interactive presentation. …

Dr. Deepak Baby is a Senior Data Scientist at KBC Bank Belgium, specializing in developing advanced data-driven solutions for banking and insurance applications. His expertise spans machine learning, speech processing, and neural network architectures. Previously, he served as an Applied Scientist at Amazon Alexa AGI, where he led research and development initiatives for speech recognition models that power Alexa services. During his tenure, he contributed to continuous learning technologies that enable Alexa models to adapt and improve over time, and participated in early-stage development of Nova Sonic LLM.
His academic background includes post-doctoral research positions at leading institutions. At Idiap Research Institute with Prof. Hervé Bourlard (May 2019 - June 2020), he advanced speech enhancement techniques using variational and adversarial auto-encoders. As a post-doctoral researcher at Ghent University with Prof. Sarah Verhulst (February 2017 - April 2019), he developed neural network-based machine learning strategies for hearing restoration, resulting in publications in Nature Machine Intelligence and Nature Communications, as well as a patent that secured $1M in ERC funding for a startup venture.
Dr. Baby holds a PhD in Electrical Engineering from KU Leuven (2016), where his dissertation focused on Non-negative Sparse Representations for Speech Enhancement and Recognition under the supervision of Prof. Hugo Van hamme. He earned his Master of Technology in Communication Engineering from IIT Bombay (2012) and his Bachelor of Technology in Electronics and Communication Engineering from College of Engineering Trivandrum (2009).






Explore distributed training techniques through this interactive presentation. …

Modern deep learning models have grown exponentially in size and complexity. …
Keras implementation can be found here. Flow-based deep generative models have …
This post is a summary of some of the main hurdles I encountered in implementing …
Variational Autoencoders (VAEs)[Kingma, et.al (2013)] let us design complex …
A hybrid approach combining convolutional neural networks with computational neuroscience to yield a real-time end-to-end model for human cochlear mechanics, including …
A CNN framework for modelling inner hair cells and auditory nerve fibers, enabling real-time simulation of the auditory periphery.
Novel approximation of noisy reverberant speech using nonnegative matrix deconvolution with decaying norm constraint for joint denoising and dereverberation.