AIM - Module 5.1 - 1 : The Deep Learning Story: From Cat Brains to AlphaFold
Quiz Questions You're working in 1970 and have built a single-layer perceptron to classify logic gate outputs. Your system correctly classifies AND, OR, and NOT gates, but completely fails on XOR inputs. Your colleague suggests the machine must be broken. What's the actual problem? 1. The potentiometers need recalibration since XOR requires higher precision weights 2. A single linear layer cannot separate the XOR function's output classes because they're not linearly separable 3. XOR requires more photodiodes in the input layer than the other logic gates 4. The training data for XOR is insufficient compared to the simpler gates You're a neuroscience researcher in the 1960s replicating the Hubel and Wiesel experiment. You've inserted electrodes into a cat's visual cortex and are showing oriented edges. Neuron cluster A fires strongly for vertical edges but not horizontal ones. Neuron cluster B fires for corners and angles. Neuron cluster C fires for complete circular shapes. How would Hubel and Wiesel classify these clusters? 1. A = hypercomplex cells, B = complex cells, C = simple cells 2. A = complex cells, B = simple cells, C = hypercomplex cells 3. A = simple cells, B = complex cells, C = hypercomplex cells 4. All three are simple cells responding to different feature types Why did the discovery of topological mapping in the cat's visual cortex have implications for designing neural network architectures? 1. It proved that random connectivity patterns are optimal for visual processing 2. It suggested that spatial relationships in input data could be preserved and exploited in network structure 3. It demonstrated that visual cortex neurons process all regions of the visual field equally 4. It showed that biological systems avoid any form of local connectivity What explains why Fukushima's neocognitron could recognize objects regardless of their position in the image? 1. Its convolutional architecture uses sliding window kernels that detect features across all spatial locations 2. It memorized all possible positions of each object during training 3. It preprocessed images to center all objects before classification 4. It used fully connected layers that inherently ignore spatial information In the Mark I Perceptron architecture described, photodiodes connect to potentiometers in a fully connected fashion. If there are 20 photodiodes and 10 potentiometer outputs, how many individual wire connections are needed for this fully connected layer? 1. 30 2. 100 3. 200 4. 20

The Scariest Chart in Electrical Engineering

A CPU Made of Atoms: IBM's Breakthrough 0.7nm Transistors

Cut LLM Cost & Latency: KV Cache, Batching, Quantization, vLLM

Visualizing transformers and attention | Talk for TNG Big Tech Day '24

Yann LeCun: World Models: Enabling the next AI revolution

AIM - Module 5.2 : CNN Architectures: Evolution of Depth, Width, and Residuals

The 3 elements of stupidity, according to philosophy | Jonny Thomson: Full Interview

How Historical Swordfight Really Looked Like

AIM - Module 5.5 : Sequence-to-Sequence: Encoder-Decoders and the Vanishing Gradient

You probably misunderstand the double slit experiment
![Yann LeCun's $1B Bet Against LLMs [Part 1]](https://i.ytimg.com/vi/kYkIdXwW2AE/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLDbV4izF3i-wxevCVIn7FJjoy1vlA)
Yann LeCun's $1B Bet Against LLMs [Part 1]

I Was NOT Ready for German Police Code 3… (American Reaction)

All 7 Dimensions Explained in Detail (From 0D to Infinity)

URGENT UPDATE - Iran War Expert: A Mass Casualty Attack Is Coming! | Robert Pape

You Don't Know Brain Eating Amoeba

Why AI Can Never Escape Turing's 1936 Proof

1: Introduction to Neural Networks and Deep Learning; Training Deep NNs

Claude Opus 5 beat every other model. Here’s the catch.

AlphaFold - The Most Useful Thing AI Has Ever Done

