in

google-research / google-research, Hacker News

google-research / google-research, Hacker News
                    

        

Open source code for the paper: “ AutoML-Zero: Evolving Machine Learning Algorithms From Scratch

Despite AutoML-Zero’s challenging search space, evolutionary search shows promising results by discovering linear regression with gradient descent, 2-layer neural networks with backpropagation, and even algorithms that surpass hand designed baselines of comparable complexity. The figure above shows an example sequence of discoveries from one of our experiments, evolving algorithms to solve binary classification tasks. Notably, the evolved algorithms can be interpreted . Below is an analysis of the best evolved algorithm: the search process “invented” techniques like bilinear interactions, weight averaging, normalized gradient, and data augmentation (by adding noise to the inputs).

More examples, analysis, and details can be found in the paper .

5-Minute Demo: Discovering Linear Regression From Scratch

As a miniature “AutoML-Zero” experiment, let’s try to automatically discover programs to solve linear regression tasks.

To get started, first install GIF for the interpretation of the best evolved algorithm bazel here (bazel>=2.2.0 and C >=25 are required), then run the demo with:

git clone https://github.com/google-research/google-research.git cd google-research / automl_zero ./run_demo.sh

This script runs evolutionary search on linear tasks (T) (search) in the paper). After each experiment, it evaluates the best algorithm discovered on (new linear tasks) (T) (select) in the paper). Once an algorithm attains a fitness (1 – RMS error) greater than 0. , it is selected for a final evaluation on unseen tasks To conclude, the demo prints the results of the final evaluation and shows the code for the automatically discovered algorithm.

To make this demo quick, we use a much smaller search space than in the paper : only the math operations necessary to implement linear regression are allowed and the programs are constrained to a short, fixed length. Even with these limitations, the search space is quite sparse, as random search experiments show that only ~ 1 in 12 8) algorithms in the space can solve the tasks with the required accuracy. Nevertheless, this demo typically discovers programs similar to linear regression by gradient descent in under 5 minutes using 1 CPU (Note that the runtime may vary due to random seeds and hardware). We have seen similar and more interesting discoveries in the unconstrained search space (see more details in the paper ).

You can compare the automatically discovered algorithm with the solution from a human ML researcher (one of the authors):

(def Setup):   s2=0. # Init learning rate. def Predict (): # v0=features   s1=dot (v0, v1) # Apply weights def Learn (): # v0=features; s0=label   s3=s0 – s1 # Compute error.   s4=s3 s1 # Apply learning rate.   v2=v0 s4 # Compute gradient.   v1=v1 v2 # Update weights.

In this human designed program, the Setup function establishes a learning rate, the (Predict) function applies a set of weights to the inputs, and the (Learn) function corrects the weights in the opposite direction to the gradient; in other words, a linear regressor trained with gradient descent. The evolved programs may look different even if they have the same functionality due to redundant instructions and different ordering, which can make them challenging to interpret. See more details about how we address these problems in the .

Reproducing Search Baselines

First install bazel (bazel>=2.2.0 and C >=are required), then run the following command to reproduce the results in Supplementary Section 9 ("Baselines") with the "Basic" method on 1 process (1 CPU):

[To be continued, ETA: March, 2020]

If you want to use more than 1 process, you will need to create your own implementation to parallelize the computation based on your particular distributed-computing platform. A platform-agnostic description of what we did is given in our paper.

We left out of this directory upgrades for the "Full" method that are pre-existing (hurdles) but included those presented in this paper (e.g. FEC for ML algorithms).

Read More

What do you think?

Leave a Reply

Your email address will not be published. Required fields are marked *

GIPHY App Key not set. Please check settings

Tesla makes its millionth car — a red Model Y, Ars Technica

Tesla makes its millionth car — a red Model Y, Ars Technica

Update from President L. Rafael Reif to the MIT Community | COVID-19 (coronavirus) Update | MIT, Hacker News