Learning to Learn

Representation Learning: Teaching Machines to See the World the Right Way Representation Learning: Teaching Machines to See the World the Right Way Here is a question worth sitting with: when you look at a cat, your brain does not process 65,536 individual pixel values and then decide "cat." Something more interesting happens — your visual system pulls out the features that matter: the pointed ears, the whiskers, the particular way it holds itself. Everything else gets discarded. You end up with a compact, meaningful summary of what you saw. Representation learning asks: can we teach machines to do the same thing? Not just to process raw data, but to automatically discover which aspects of that data actually matter — and compress everything into a form that is useful for reasoning, comparing, and deciding. This turns out to be one of the most important ideas in modern AI. Every large language model, every image classifier, every system that lets ...

MobileNet Backbone Architecture Versions MobileNet Backbone Versions: Designing Efficient CNNs for Real-World Deployment MobileNet is a family of efficient convolutional neural networks designed for real-time inference on resource-constrained devices such as smartphones, drones, and embedded IoT hardware. Over multiple versions, the MobileNet family introduced progressively refined design innovations — from depthwise separable convolutions to neural architecture search and transformer-style attention — while keeping computation minimal. Rather than treating MobileNet as a single model, this blog explores each version's motivation , the specific problem it addressed , and the architectural innovations it introduced to push the accuracy-efficiency frontier. 1. Why MobileNet? Motivation and Core Problem Standard convolutional networks like VGG and ResNet are accurate but computationally heavy. Deploying them on edge devices with limited memory, power, ...

Variational Inference: A Mathematical Explanation Variational Inference (VI) Variational Inference (VI) is a mathematical framework for approximating complex posterior distributions in probabilistic models. Instead of sampling (as in Monte Carlo methods), VI transforms inference into an optimization problem, making it scalable and efficient for deep learning. 1. Problem Setup We consider a probabilistic model with observed variables \( x \) and latent variables \( z \). The joint distribution is defined as: $$ p_\theta(x, z) = p_\theta(x \mid z)\, p_\theta(z) $$ Notation: \( x \): observed data \( z \): latent (hidden) variable \( p_\theta(x \mid z) \): likelihood (decoder) \( p_\theta(z) \): prior distribution \( \theta \): model parameters The goal is to find parameters that maximize the marginal likelihood of the observed data: $$ p_\theta(x) = \int p_\theta(x, z)\, dz $$ However, this integral is often intractable due to hig...