MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·4h ago · 8 min readText Recognition: How Machines "Read" a String of Digits from an ImageTable of Contents Why This Matters Guess First What Is Text Recognition (OCR) Why CNN Alone Isn't Enough LSTM's Job: Understanding Sequence The Alignment Problem and Why CTC Loss Exists From Th00
MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·1h ago · 7 min readBefore a Model Can Read a Sentence, It Has to Recognize One Letter FirstTable of Contents Why This Post Exists Guess First How This Differs from Text Recognition (CRNN + CTC) Dataset: A-Z Handwritten Alphabets Architecture and Training Setup Training Results Per-Cl00
TTharunintharunpandya.hashnode.dev·5h ago · 7 min readThe Sigmoid Function: Squashing the World Into 0 and 1You open YouTube. You haven't searched for anything, yet the very first video looks strangely interesting. You click. Imagine a simplified recommendation model working behind the scenes, quietly askin00
LMLakshmi Mahabaleshwarainfreecodecamp.org·13h ago · 21 min readHow to Turn a RECIST Line into a 3D Tumor Segmentation MaskA radiologist can mark a tumor on a CT scan by drawing a straight line across it. Creating a complete 3D segmentation requires outlining the tumor across the slices where it appears, which is more tim00
SPSamu Physicistinsamu-physicist.hashnode.dev·2d ago · 15 min readTPU-Friendly Kolmogorov-Arnold Networks for Neural QMC: A 330x Compression of FermiNet Using JAX[TL;DR] What if you could run Kolmogorov-Arnold Networks (KANs) efficiently on TPUs without suffering from memory fragmentation and conditional branching? Note for ML Engineers: While this project tac00
LPLakshya Padhanintheainotebooklakshya.hashnode.dev·3d ago · 14 min readAttention Is All You Need: The Paper That Changed Modern AIAn accessible deep dive into the 2017 research paper that introduced the Transformer architecture In 2017, a research paper with a surprisingly confident title — “Attention Is All You Need” — introduc10
MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·3d ago · 11 min readSingle Object Detection: Why the Bounding Box Wasn't RightTable of Contents Quick Recap Second Iteration: Augmentation and Partial Freeze Testing the Model on Real-World Photos Plot Twist: A Box That Clips the Object Third Iteration: Split Head and IoU 00
MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·3d ago · 8 min readSingle Object Detection: Training and Serving with FastAPITable of Contents Quick Recap Setting Up the Dataset and Model IoU: Measuring How Good a Prediction Is The Full Training Loop First Training Results Serving the Model with FastAPI Cheat Sheet 00
MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·3d ago · 7 min readSingle Object Detection: Core Concepts and ArchitectureTable of Contents Why Single Object Detection What Single Object Detection Actually Is How It Works: From Pixels to Bounding Box Loss Function: Juggling Three Goals at Once Cheat Sheet Up Next i00
TTech-Aarvamintechaarvam.hashnode.dev·3d ago · 14 min readPuzzle Solution Revealed - Transformer: Need for Position EmbeddingIntroduce position embedding, residuals and layers ▶ Watch the 7-minute walkthrough import torch torch.set_printoptions(precision=2, sci_mode=False, linewidth=160) In this notebook we are extending00
