ML.pdf

Machine Learning talk at iOS Jam Vienna 2018

Transcript

1. Machine learning
3. Machine learning gives computers the ability to learn without being explicitly programmed.
4. Neural networks Computing systems inspired (loosely) by the brain
5. Neuron - unit of computation ŷ x1 x2 x3 x4 x - real number non-linear function of the sum of x
6. Neuron network ŷ Input layer Hidden layer Output layer
7. Training (deep learning) ŷ Input layer Hidden layer Output layer Loss function Loss function adjusts connections weights as learning proceeds
8. Transfer learning Classifier Bottleneck Retrain
9. Supervised learning A B C Classification Regression
10. Clustering A B D C Unsupervised learning
11. Frameworks
12. Manchester United Not Manchester United
13. Model creation
14. Bounding boxes Origin 575, 320 width 410 height 450
15. Turi Create data = tc.image_analysis.load_images('train', with_path=True) data = data.sort(['path']) data['annotations'] = [ [{'coordinates': {'height': 450, 'width': 410, 'x': 575, 'y': 320}, 'label': 'manch'}], #0 [{'coordinates': {'height': 271, 'width': 261, 'x': 280, 'y': 176}, 'label': 'manch'}], #1 [{'coordinates': {'height': 177, 'width': 177, 'x': 254, 'y': 250}, 'label': 'manch'}], #2 … [{'coordinates': {'height': 400, 'width': 380, 'x': 360, 'y': 235}, 'label': 'manch'}], #31 [{'coordinates': {'height': 190, 'width': 190, 'x': 250, 'y': 205}, 'label': 'manch'}], #32 [{'coordinates': {'height': 170, 'width': 170, 'x': 350, 'y': 620}, 'label': 'manch'}] #33 ] model = tc.object_detector.create(data) model.export_coreml('manchester.mlmodel') import turicreate as tc
16. Core ML framework
17. Core ML request let request = VNCoreMLRequest(model: model, completionHandler: { [weak self] request, error in self!?.processClassifications(for: request, error: error) }) private var requests = [VNCoreMLRequest]() let mlmodel = Manchester().model let model = try VNCoreMLModel(for: mlmodel) request.imageCropAndScaleOption = .scaleFill requests.append(request)
18. Core ML request handler let handler = VNImageRequestHandler(cvPixelBuffer: pixelBuffer, orientation: CGImagePropertyOrientation(rawValue: UInt32(ex))!, options: requestOptions) guard let pixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else { return } handler.perform(requests)
20. •course.fast.ai •coursera.org/learn/machine-learning with Andrew Ng •Swift by Sundell 22: "A path for learning” •lynda.com iOS App Development: Core ML •github.com/blob8129/MUnotMU Resources