Monthly Archives: January 2016

On Singularities

A computer, once again, outperformed a human in a highly specific task, this time around the game of Go, using, in part, a recent (well, not that recent, but also not “traditional”) AI technique known as Deep Learning.

The media, once again, made a splash, and some critics were quick to dismiss the feat by pointing to the limitations of Deep Learning (the wining algorithm also used more traditional AI methods).

Of course, the people on the forefront of Deep Learning know better than anyone about its limitations — they’re simply more faithful in it than others. Deep down (no pun intended), they probably don’t like such splashy news either, because it raises expectations, but we all understand the importance of advertisement (we live in a social, political world).

If history is of any guidance, the current hype will pass, as have many other AI hypes. It is not impossible that general AI will happen. Singularities do happen: this universe, self-replication, self-consciousness. But they seem to occur only every billion years or so.

Hence, in the big scheme of things, the last singularity happened just “yesterday,” and we will have to wait a whole lot for the next. Current AI progresses are admirable, and important, but as a society, we have to learn to look at them for what they really are: incremental steps.

Related: On the Higgs Boson Hysteria

TensorFlow 101

There’s a sort of “gold rush” between Machine Learning toolkits to grab the attention of developers. Caffe, Torch, Theano, and now TensorFlow, are only some of the competitors. Which one to choose?

Hard to know for sure. There are the usuall technical trade-offs (see, but for the user, besides technical capabilities, often times the choice comes down to which one has the best documentation (i.e., which one is easier to use).

So far, given the power of it’s sponsor, TensorFlow seems to be the one with a more serious approach to documentation. Still, the MNIST and CNN tutorials could be simpler.

Introducing: TensorFlow 101.

This project has two main files ( and, and two sample datasets (subsets of the MNIST and CIFAR10 databases). The Python routines are modified from the “MNIST For ML Beginners” and “Deep MNIST for Experts” (from

The main diference is that and contain code to read and build train/test sets from regular image files, and therefore can be more easily deployed to other databases (which, ultimately, is the goal of the user). Notice that the folders MNIST and CIFAR10 are organized in subfolders Train and Test, and in these subfolders each class has a separate folder (0, 1, etc).

Therefore, one simple way to deploy and to your own custom database is to organize your database in the same hierarchy as MNIST and CIFAR10 in this project, and modify the variable “path” on the .py routines to point to your dataset. Notice that your dataset doesn’t have to have 10 classes; however, all images in the provided sample datasets are grayscale and have size 28×28, hence non-trivial modifications to the code should be performed in order to deal with other types of images.

See also: Udacity Deep Learning Course.