From course, I learned Evolutional Computing. Evolutionary computing (EC) population method Used for problems where the task is to maximize some measure of success Same family of problems as in RL, but very different methods Methods inspired by genetics, natur…

Hello everyone! 这篇文章将介绍Dice coefficient以及其实现 Dice coefficient 是 Lee R. Dice 在1945年为评估生物种群提出的一种度量方法[1]。后来不同领域的学者都将其引入到自己的专业。这里，我将介绍Dice codfficient 在图像分割领域作为评价指标的理解与实现。 图像分割包含有语义分割、示例分割。在语义分割中，我们需要对给定的图片做…

Let us firstly revisit completive learning and SOFM A node position (in the input space) is its weight vector, i.e. nodes move around in the same space as the data. Purpose of unsupervised learning is (usually) to cluster for classification, or for modelling d…

In this note, we are going talk about Deep Learning very briefly. RBFNs are feedforward networks, where: The hidden nodes (the RBF nodes) compute distances, instead of weighted sums. The hidden layer activation functions are Gaussians (or similar), instead of …

To learn from unlabeled data (only inputs, no target information). Most often for classification. Requires that class membership can be detected by structural properties (features) in the data, and that the learning system can be fund these features. Following…

ex) Tic-Tac-Toe Assign value to each board position (state): +1 if the state is a winning terminal state. 0 if the state is a losing terminal state +0.5 to all other states (terminal and non–terminal) play many games After each move, adjust the value of the pr…

From today, we are going to introduce reinforcement learning(RL). Forget ANNs for a while - RL is a field in its own right, independent from ANNs. However, as we will see, ANNs are sometimes used in RL. Definition: Learning by interaction with an environment, …

In this course, we learn more on extensions. and prior knowledge. We want to minimize the number of hidden nodes(=> less overfitting) The approaches: start with a large networks and prune. start with a small networks and grow. There are two example. Let eac…

Let's continue talk about overtraining. The number of training set samples should be much larger than the number of weights roughly for ​ \(N^2\)(N = the number of weights). minimize the number of nodes and layers(weights) noise injection and noise to the…

in order to decide by how much of the result blamed or credited to a particular weight. We need sigmoid function. Sigmoid function = any S-shape function. examples. logistical function \( y=f(s)=\frac{1}{1+e^{- \lambda s}} \), \(\lambda\) decides slope. When \…