symbolic math library used for creating neural networks and deep learning models. TensorFlow is very flexible and the primary benefit is distributed computing. CNTK is deep learning framework developed by data preparation phase of machine learning.  Image processing: Provides functions to convert images into NumPy array suitable for machine learning. We can use it in data preparation phase of machine Keras supports each concept. Input shape In machine learning, all type of input data like text, images or videos will be first converted into array of numbers and then feed into the algorithm. Input

transformed distribution improves the model quality and performance. For instance, a dataset of cat images would likely have the cats positioned at various angles. It would make sense to have rotational transformation during the training process. Tensorflow comes bundled with the ImageDataGenerator which can transform images during the training process, thus eliminating the need to generate samples beforehand. Next, we will the holes. Figure 3-6: Image Transformations. The source image (center) is taken from Google Open Images Dataset V6. It is authored by Mike Baird and is licensed under CC BY 2.0. The image is resized to

losing some information as a trade off. It is especially applicable for multimedia (audio, video, images) data,, where it is likely that either humans who will consume the information will not notice the quality image of a cat. The cat on the right is a lower quality compressed image. Source Both the cat images in figure 2-2 might serve their purpose equally well, but the compressed image is an order of magnitude have a task for you. You have been handed the charge of the Mars Rover! The rover is transmitting images back to earth. However, transmission costs make it infeasible to send the original image. Can we

dataset This guide uses the Fashion MNIST dataset which contains 70,000 grayscale images in 10 categories. The images show individual articles of clothing at low resolution (28 by 28 pixels). https://github fashion_mnist (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data() Preprocess data import matplotlib.pyplot as plt plt.figure() plt.imshow(train_images[0]) plt.colorbar() colorbar() plt.grid(False) plt.show() import matplotlib.pyplot as plt plt.figure() plt.imshow(train_images[1]) plt.colorbar() plt.grid(False) plt.show() Preprocess data class_names = [ 'T-shirt/top'

step=step) # 可视化测试用的图片，设置最多可视化 9 张图片 tf.summary.image("val-onebyone-images:", val_images, max_outputs=9, step=step) 运行模型程序，相应的数据将实时写入到指定文件目录中。 8.7.2 浏览器端 在运行程序时，监控数据被写入到指 SCALARS、图片可视化页面 IMAGES 等。对于这个例子，我们需要监控的训练误差和测 试准确率为标量类型数据，它的曲线在 SCALARS 页面可以查看，如图 8.4、图 8.5 所 示。 预览版202112 第 8 章 PyTorch 网络模型 16 图 8.4 训练误差曲线 图 8.5 训练准确率曲线 在 IMAGES 页面，可以查看每个 Step 的图片可视化效果，如图 step=step) # 可视化测试用的图片，设置最多可视化 9 张图片 tf.summary.image("val-onebyone-images:", val_images, max_outputs=9, step=step) 运行模型程序，相应的数据将实时写入到指定文件目录中。 8.7.2 浏览器端 在运行程序时，监控数据被写入到指

Recognizing hand-written words P: Percentage of words correctly classified E: Database of human-labeled images of handwritten words Feng Li (SDU) Overview September 6, 2023 10 / 57 What is Machine Learning using vision sensors P: Average distance traveled before a human-judged error E: A sequence of images and steering commands recorded while ob- serving a human driver Feng Li (SDU) Overview September expression levels for predicting tumor type, age and income for predicting amount spent, the features of images with known semantics Feng Li (SDU) Overview September 6, 2023 24 / 57 Classification and Regression

For other domains like images, it is not practical to store all possible inputs, or even the top K frequent inputs. There isn’t a vocabulary as such to begin with, since images with a single pixel of difference this approach by using data augmentation to create similar inputs without needing labels for the images. Once the model generating the representation is trained, it can then be frozen and a new classification figure 4-22. We are going to use the Oxford-IIIT Pet dataset30 for training purposes. It contains images and segmentation masks for 37 pet categories. We will train two models using regular convolutions

应⽤用：分类训练集与验证集划分 https://www.pinlandata.com/rp2k_dataset Categoried Stats # SKUs # Train images # Test images # Total Images # Images/SKUs Low-Temperature Dairy(低温乳制品) 198 67666 7612 75278 380.19 Liquor(白酒) 167

fine-tuning stages. In the figure we demonstrate pre-training with a large unlabeled dataset of animal images. The pre-trained model is then fine-tuned for downstream tasks, for example object detection for org/I05-5002. Figure 6-1: Pre-training and fine-tuning stages. With an example of a large unlabeled animal images dataset which is used for pre-training. The pre-trained model is then used for fine-tuning for downstream Therefore, we can simply use e-books, Wikipedia and other sources for NLU related models, and web images & videos for computer vision models. We can then construct the final dataset for the pretext task

Squeeze/unsqueeze ▪ Transpose/t/permute ▪ Expand/repeat 2 View reshape ▪ Lost dim information 3 Flexible but prone to corrupt 4 Squeeze v.s. unsqueeze 5 unsqueeze [-a.dim()-1, a.dim()+1) [-5, 5)