@objcio

Best practices and advanced techniques for iOS and macOS development

objc.io
Joined May 2013
  • Tweets2.2K
  • Following3
  • Followers36.3K
Swift Talk 499 Split Pane Animations (Part 2) We profile the smoothness of our animations, using the Animations Hitches instrument with signposts to help us identify the right sections. talk.objc.io/episodes/S01E49…
1
9
2,356
Swift Talk 498 Split Pane Animations (Part 1) We build a split-pane view that can divide panes dynamically, either horizontally or vertically, with smooth animations for adding or resizing. This episode is free to watch, enjoy! 😊 talk.objc.io/episodes/S01E49…

Swift Talk 498 — Split Pane Animations (Part 1)

We build a basic split pane view as a basis for profiling animation smoothness.

3
26
3,368
Swift Talk 497 Building a Language Model: MLX We use Apple's MLX framework to make our neural network much faster, with shorter, cleaner code using built-in convenience functions. talk.objc.io/episodes/S01E49…

Swift Talk 497 — Building a Language Model: MLX

We use Apple's MLX framework to train our neural net with a massive performance increase.

1
21
3,617
Swift Talk 496 Building a Language Model: Fixing Backpropagation We fix two mistakes with our reimplementation of @karpathy’s micrograd, switching from a recursive to an iterative, topologically sorted approach to backpropagation. talk.objc.io/episodes/S01E49…

Swift Talk 496 — Building a Language Model: Fixing Backpropagation

We switch from a recursive to a topologically sorted algorithm, to compute the gradient of the cost function.

6
1,698
Swift Talk 495 Building a Language Model: Gradient Descent We make the forward pass differentiable, building a graph that backprop walks in reverse to compute gradients. Gradient descent uses those to train the network, updating weights to reduce loss. talk.objc.io/episodes/S01E49…

Swift Talk 495 — Building a Language Model: Gradient Descent

We lay the groundwork for backpropagation in our neural net.

1
1,185
Swift Talk 494 Computing the Loss Function We finish the neural network’s forward computation, turning raw inputs into probability distributions. Comparing those with the targets gives us a loss value: a measure of how good the predictions are. talk.objc.io/episodes/S01E49…

Swift Talk 494 — Computing the Loss Function

We continue to work on the forward computation of the neural net by implementing the loss function.

3
1,202
Swift Talk 493 Digrams with Neural Nets We start building a simple neural net to recognise digrams. Swift doesn’t have built-in types for things like matrices, so we create our own. talk.objc.io/episodes/S01E49…

Swift Talk 493 — Digrams with Neural Nets

We start to implement digrams using the neural net implementation from last episode.

10
1,391
Swift Talk 492 Building a Language Model: Neural Nets We explore neural nets, the foundation for modern machine learning. Our first small step is backpropagating through a computation graph, inspired by @karpathy’s Micrograd. talk.objc.io/episodes/S01E49…

Swift Talk 492 — Building a Language Model: Neural Nets

We investigate neural nets, which form the foundation for modern machine learning.

12
1,567
Swift Talk 491 Building a Language Model: Tokenization We tokenize our text with byte-pair encoding, merging frequent character pairs to build a new vocabulary of tokens. talk.objc.io/episodes/S01E49…

Swift Talk 491 — Building a Language Model: Tokenization

We implement tokenization using the byte-pair encoding algorithm.

1
22
2,937
Swift Talk 490 Building a Language Model: N-grams We build a simple n-gram implementation to better understand language models, training a complete—if slightly useless—model on a single book. This episode is free to watch, enjoy! 😊 talk.objc.io/episodes/S01E49…

Swift Talk 490 — Building a Language Model: N-grams

We implement n-grams as a first step to better understand language models.

1
3
16
2,179
Load more