Deep Learning may be the next great frontier in computer development, essentially taking the concept of artificial intelligence and turning it into a reality. While AI has been around for years, many developers believe taking it to the next level is only a matter of time. That’s why so many are excited by Facebook’s recent announcement at the International Conference on Machine Learning (ICML) back in June. This announcement revealed the launch of Torchnet, an open source software toolkit meant to build upon the already existing Torch, which itself is an open source library for a deep machine learning framework. The reveal of a deep learning framework isn’t exactly new, but Facebook’s approach is a different strategy. The goal will be to streamline deep learning advances among developers, with the hope that artificial intelligence takes that next step sooner rather than later.
Facebook is certainly no stranger to deep learning and artificial intelligence. Facebook makes extensive use of advanced algorithms in many of its functions. From determining what users see on their timelines to the latest in facial recognition software for identifying people in photos, Facebook has come a long way from the time people would simply post updates on what they were doing at the moment. In many respects, Facebook has turned into the full social media experience, incorporating numerous elements of other networks in effort to stay on top of the competition at every turn. To see Facebook go all in on deep learning development is far from a major surprise.
The Torchnet announcement comes as other big names in the tech world have released their own deep learning frameworks. But whereas the like of Google, Amazon, and Microsoft released completely new frameworks, Facebook has chosen to build their framework off of an already established open source library -- Torch -- one that many developers are familiar with. The idea is to provide developers with the ability of rapid experimentation. Facebook itself describes Torchnet’s design to be comparable to Lego building blocks, basically allowing developers to snap individual coded chunks together easily or replace them entirely depending on what they’re trying to do. In this way, different coding variants can be tested and re-tested with relative ease, helping researchers find new and better ways to draw closer to more advanced deep learning techniques.
Many developers consider Torchnet to be a definite step up from Torch. One of the improvements cited is the fact that programmers won’t have to always code from scratch. As is the case with other frameworks and Torch, coders often have to cover the same ground and code the same things with every experiment. Torchnet gives these programmers the guidelines to streamline this process along with boilerplate code that plays a large role in helping developers make greater progress more quickly. Not only does this streamline coding, but it leads to less work for programmers and less of a chance for bugs to be introduced. Whether used for big data analytics research or machine learning optimization, the strengths are readily apparent.
Since Torch is already used by many programmers for things like metric learning and neural networking, the transition into using Torchnet should be a simple one, something Facebook is no doubt counting on. Not only does this mean developers can take what they’ve learned from Torch and easily convert in for the new Torchnet community, but they’re findings can then be applied to other frameworks. Abstractions can be implemented for frameworks like TensorFlow and Caffe, making Torchnet highly versatile and invaluable in deep learning development. With these tools, greater collaboration between programmers is expected, and with all the best minds working on artificial intelligence, there’s no reason to think greater strides won’t be made in the coming years.
It still isn’t known specifically how Torchnet is used within the Facebook organization, but the potential it provides in other areas is undeniable. As more organizations move to discover more insights in machine learning, software defined storage, and big data, Torchnet is likely to play a pivotal role in that research. The tools are now readily available, and programmers and developers can now pick the approach that best suits them and what they hope to accomplish.
This post first appeared on CTOvision.com - Context For The CTO, CIO, CISO And, please read the originial post: here