Machine Learning for Engineering & Science Applications | IIT Madras - #95 Variational Auto Encoders (VAE) | Machine Learning for Engineering & Science Applications

Welcome to 'Machine Learning for Engineering & Science Applications' course ! Want to uncover the hidden structure of your data and generate meaningful new samples? This lecture takes you on a journey through Variational Autoencoders (VAEs). We'll start with a refresher on Autoencoders, neural networks that learn to compress and reconstruct data. But VAEs go further, introducing a probabilistic twist to the latent space, the compressed representation of your data. With an encoder mapping data to latent space parameters and a decoder generating samples from these parameters, VAEs enable us to model the distribution of our data. We'll explore the role of KL Divergence in regularizing the latent space and understand how this leads to meaningful representations and smooth interpolation between generated samples. NPTEL Courses permit certifications that can be used for Course Credits in Indian Universities as per the UGC and AICTE notifications. To understand various certification options for this course, please visit https://nptel.ac.in/courses/106106198 #VariationalAutoencoders #VAEs #Autoencoders #LatentSpace #LatentVector #Encoder #Decoder #KLDivergence #ReconstructionLoss #Regularizer #GaussianDistribution