tf overview

Tensorflow is a computer software, namely, software for implementing a computer programming language and computer software development tool for use in the fields of artificial intelligence, deep learning, high performance computing, distributed computing, virtualization and machine learning; computer software libraries for use in general purpose computation, manipulation of collections of data, data transformation, input/output, communications, graphics display, modeling and testing. Well that explanation was long and pulled from the USPAT trademark website.

So what is TensorFlow in laymans terms? A software that can distinguish cats and dogs from pictures or estimate how many pigeons will land on a tree at any given time (and much more). According to Google TensorFlow is a second generation deep learning algorithm.

● 1st generation: DistBelief [Dean et al., NIPS 2012]
● 2nd generation: TensorFlow (published Nov 2015)

TensorFlow itself is written in C++ and currently has programming APIs for C++ and Python. The TF creators hope that the global community provides more APIs for a number of programming and scripting languages (JAVA, C#, R and others). TF is currently neither the fastest nor the most accurate algorithm or framework, but potentially can be very fast and scale very well with multiple GPUs (graphic cards).

2016: Whats new? Well a much improved TF documentation with many more examples and a number of code wrappers such as prettytensor and skflow a simplified TF interface.

---

Links

• Deep Learning - Large-Scale Deep Learning for Intelligent Computer Systems by Jeff Dean