Deep learning trends and an example
The Spark Summit 2016 took place on June 6–8 in San Francisco and it was a sold out event with more than 2,500 attendees. Not surprisingly, deep learning (DL) and artificial intelligence (AI) were the main dishes of the conference. On day one, most of the keynotes were on how DL and AI are making the world better. Don’t you think it’s amazing that you can teach a computer to distinguish between an image of a cat from an image of a dog? I do! And the mentioned example is nothing compared to others fantastic examples that were presented at the Summit. Based on google searches, starting around 2014, both terms Apache Spark and Deep Learning have had a dramatic increase.
Jeff Dean, head of Google’s brain team, talked about how DL is used to verbally describe an image. Imagine a blind person using an app to understand an image without the help of other person! Jeff also talked about other use-cases where DL is useful like speech recognition and email smart reply, among others.
Andrew Ng, chief scientist at Baidu and co-founder of Coursera, compared AI models with rockets: artificial neural networks to their engine and data to its fuel. At Baidu, DL and AI are being applied to train models for autonomous driving, fraud and malware detection, among other use-cases. Neural networks (NN) need more data than traditional algorithms, especially deep neural networks. The gain of NN algorithms trained with large amounts of data is in the quality of your predictions at a cost of more computational power (therefore the popularity of GPU’s used for training NN’s). Find the slide shown and more information about all the very interesting talks at the Spark Summit here.
Hopefully I have convinced you that DL and AI is quiet something to look at. This is why I build a notebook on Data Science Experience to run a very well-known and simple DL example for classifying handwritten digits. Please check my notebook here.
The best way to contact me for questions, feedback or just to say hi is @castanan.
Originally published at datascience.ibm.com on June 29, 2016.