Finally Machines Can Learn as People!!!

Before we move forward with this article, it is essential that the the word 'Algorithm' be defined accurately. (An Algorithm is a prescribed set of well-defined rules or instructions, and finite ordered that enables activity through successive steps that do not generate doubts to who should perform this activity. Given an initial state and entry, following successive steps reaches a final state and a solution is obtained. The algorithms are the subject of study of algorithmics - see Figure 1 -) Once again, the interesting technique known as an algorithm is being employed to teach machine to learn and recognize objects. This will result in expanding and better understanding the human cognitive system.



Figure 1 Flow charts are used to graphically represent algorithms.

The famous Massachusetts Institute of Technology recently revealed the results of a new research in the field of Artificial Intelligence in the form of documents. They reported progress on the so-called 'object-recognition system’, which is beginning to show that they have made the correct manuever. This is especially important in the more complex and full understanding of algorithms used within the field of facial recognition, which is successfully being used by the Facebook company.

We must understand that this new model has the same design of how human beings recognize their suroundings. The new system uses the similarity of patterns among millions of photos to be compared with the 'original' picture. However, the human memory undoubtedly does not work in this way. A man needs to only see two or three pictures of an object in order to identify the details on it, and with relative accuracy.

This program has been under investigation for the past couple of years: led by Tomaso Poggio, Institute for Brain Research at MIT. Poggio initialized investigations of this new computational model of visual representation and intended to reflect how the brain actually processes visual information. It was reported that these developments will be explained in detail in the next issue of ‘Theoretical Computer Science ' journal. The research shows that a learning machine system based on that model could discriminate or discern objects, and have used a few examples to verify this statement. Consequently, the researchers are aiming towards achieving more reliable results.

This research along with similar research appeared in the “PLOS Computational Biology’ journal. The research depicted that several aspects of the model are in empirical concordance of how the brain actually works.

According to Professor Poggio, '' If you observe a picture of a face, at a distance, and then when you observe the picture again but from a different distance, then you note that the picture is very different, so simply do not work". Professor Eugene McDermott of the Sciences Department of Brain and Cognitive Sciences at MIT said: "To resolve this issue, whether a large number of examples is needed, we need to see the face, not only in the same position, but at all possible positions, or to know also that it is a representation invariant of the object positions. "

An invariant representation of an object is one that is immune or not affected to differences such as size, location, and rotation in the lane. Computational Researchers in the field of vision have proposed several techniques in the invariant representation of objects, but the group of Poggio has a greater challenge. The group must find an invariant representation that is consistent with what we know about the incredible machinery of the human brain.



From a neural point view, we must understand that the nerve cells (neurons) are thin and long, and have an incredible potential of ramifications (dendrites). In the cerebral cortex, where the visual 'processing' occurs, each neuron has about 10,000 branches in each one of its ends (axons).
Therefore, two cortical neurons (cerebral cortex) communicate with each other through the 10,000 different chemical bonds, known as "synapses'. Each synapse has its own 'weight', a factor by which the strength of an incoming signal is multiplied. The signals cross all 10,000 synapses, then are added in full to the neuron body (soma). Subsequently, the patterns of electrical activity stimulation changes, and modifies as time passes. The same occurs with the weights of synapses, which is the mechanism by which memories and habits become entrenched or fixed.

Mathematically speaking, a major operation in the branch of mathematics known as linear algebra is the key point in this process. This operation has two sequences of numbers - or vectors -, and they multiply their elements in an orderly manner. Subsequently, adding all the results would be a single number. In the cerebral cortex, the output of a single neural circuit could be considered as the product of two vectors consisting of 10,000 variables. That’s a very large calculation that each neuron in the brain can do at a ‘stroke’.

Poggio’s group then developed an invariant representation of objects that is based on dot-products. Suppose that you make a little digital movie of an object rotating 360 degrees in a plane. Now let us take 24 frames, each depicting the object as rotated a little bit further than it was in the last one. You then store the movie as a sequence of 24 stills.

Suppose next that you’re presented with a digital image of an unfamiliar object. Because the image can be interpreted as a string of numbers describing the color values of pixels, - a vector - you can calculate its dot-product with each of the stills from your movie and store that sequence of 24 numbers.



Now, if you’re presented with an image of the same object rotated about 90 degrees, and you calculate its dot-product with your sequence of stills, you’ll get the same 24 numbers. They won’t be in the same order. Rather, what was once the dot-product with the first still will now be the dot-product with the sixth. Ultimately they will be the same numbers though.

That list of numbers acquired is then used as a representation of the new object that is invariant to rotation. Similar sequences of stills, which depict an object at various sizes or at various locations around the frame, will yield sequences of dot-products that are invariant to size and location.

In their newest paper, Poggio and his colleagues — first author Fabio Anselmi, a postdoc in Poggio’s group; Joel Leibo, a research affiliate at the McGovern Institute and a research scientist at Google DeepMind; Lorenzo Rosasco, a visiting professor in the Department of Brain and Cognitive Science; and Jim Mutch and Andrea Tacchetti, graduate students in Poggio’s group — demonstrate that, if the goal is to produce an object representation invariant to rotation, size, and location, then the ideal template is a set of images known as Gabor filters. And Gabor filters, it turns out, are known to offer a good description of the image-processing operations performed by the so-called “simple cells” in the visual cortex.

While this technique works well for visual transformations within a plane, it doesn’t work as well for rotation in three dimensions. The dot-product between a new image and that of a car for example, would be very different from the dot-product of the same image and that of a car seen from the side.

But Poggio’s group has shown that if the template of still images depicts an object of the same type as the new object, dot-products will still yield adequately invariant descriptions. This observation resonates with recent research by MIT’s Nancy Kanwisher and others, indicating that the visual cortex has regions specialized for recognizing particular classes of objects, such as faces or bodies.



In the work described in the PLOS Computational Biology journal, Poggio and his colleagues — Leibo, Anselmi, and Qianli Liao, a graduate student in electrical engineering and computer science — built a computer system that assembled a set of still images and used the dot-product algorithm to learn to classify thousands of random objects.

For each of the object classes that the system learned, it produced a set of templates that predicted the size and variance of the regions in the human visual cortex devoted to the corresponding classes. That suggests, but the researchers argue, that the brain and their system may be doing something similar.

The researchers’ invariance hypothesis is “a powerful approach to bridge the large gap between contemporary machine learning, with its emphasis on millions of labeled examples, and the primate visual system that in many instances can learn from a single example,” says Christof Koch, a professor of biology and engineering at Caltech and chief scientific officer of the Allen Institute for Brain Science. “This sort of elegant mathematical framework will be necessary if we are to understand existing natural intelligent systems, on the road to building powerful artificial systems.”



The researchers’ work was sponsored, in part, by MIT’s Center for Brains, Minds, and Machines, and was funded by the National Science Foundation and directed by Poggio.
Source:
Larry Hardesty | MIT News Office
December 31, 2015

YOUR INPUT IS MUCH APPRECIATED! LEAVE YOUR COMMENT BELOW.

Share this: