When it comes to Generative AI, we often hear the terms encoding and decoding. A key term that we don’t often hear associated with these encoding and decoding is latent space.
What is common to many Generative AI models is the concept of encoding the training dataset into a latent space. Then, you sample from that latent space and decode the point back to the original domain. This encoder-decoder technique attempts to transform the highly nonlinear complexities on which the data resides (for example, a pixel space) into a simplified latent space. This latent space can then be sampled from in such a way that any point in the latent space is essentially a representation of a well-formed image.
Watch this episode of Gen AI Bytes to gain a better understanding of Representation Learning.
Kumar Singh is the founder of Designed Analytics LLC, which is focused on helping organizations explore how to leverage data and analytics to effectively compete, thrive and innovate.
Kumar has over a decade of hands-on experience in supply chain and operations analytics. With over a decade of industry experience, he has worked across multiple industries, helping companies set up analytics centers of excellence.
Post his industry experience, Kumar also did a stint in external consulting as a data science consultant with Boston Consulting Group (BCG). Kumar holds an MSc. in AI from Liverpool John Moores University, a Masters in Supply Chain from The Ross School of Business at The University of Michigan, Ann Arbor, an MBA in Operations Management from IIT Roorkee, India, and an undergraduate degree in Electrical engineering.
Kumar is ASCM CPIM, CSCP, and CLTD certified and holds the PMP certification from PMI. He is also an AWS-certified Machine Learning Specialist and Microsoft certified in the Azure IoT developer specialty.
Designed Analytics Blog 