New answers tagged

1

It will recover the encrypted inputs. The algorithm starts with dummy data and dummy label, and then iteratively optimizes the dummy gradients close as to the original, also makes the dummy data close to the real training data: $$\mathbf{x}^{\prime *}, \mathbf{y}^{\prime *}=\underset{\mathbf{x}^{\prime}, \mathbf{y}^{\prime}}{\arg \min }\left\|\nabla W^{\...


2

To obtain guarantees of convergence for Q table values, you need to decay the learning rate, $\alpha$, at a suitable rate. Too fast and convergence will be to inaccurate values. Too slow and convergence never happens. For sticking with theoretical guarantees, the learning rate decay process should generally follow the rule that $\sum_t \alpha_t = \infty$ but ...


2

I'm not sure if I understood your question correctly, but here's my take anyway. So, PCA is a technique that you can apply to data to reduce the number of features. In return, (i) this can speed-up training, as there are less features to do computation with, (ii) and can prevent overfitting, as you lose some information on your data. To detect overfitting, ...


3

The vast majority of neural networks are now trained on graphics processing units (GPUs) or specialised accelerator hardware such as tensor processing units (TPUs). In Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm, Silver et al. say that the training process involved 5,000 first-generation TPUs generating self-play ...


0

I don't actually understand your question, but if your data is completely arbitrary, there is nothing to predict, it have no patterns to recognize or something like that. But if you say that you are working with time-series data and it have some patterns, then you could start by trying to implement just the forward propagation of a simple RNN. I am really ...


Top 50 recent answers are included