Breakthroughs in Deep Learning


Deep learning on neural networks has been with us from as early as the 1960s. However, several advancements in the field such as the neuron layers types, the way you train the neural network, the amount of data available to train with as well as the processing power, have now made deep learning more practical to use. 

One of the first major breakthroughs came around 2012. At that time the academic research community struggled with the challenge of classifying millions of images into a thousand different categories. Alex Krizhevsky managed to do this through some very clever engineering and by building an efficient neural network, which surpassed any other technique. 

It was so disruptive that it resulted in an arms race between the academic research community and private companies. While Alex had to write specific code to run on GPUs, nowadays dedicated AI processors are built in order to support the type of layers Alex was advocating. Since then, we’ve witnessed several breakthroughs in this scientific area and the pace and magnitude of the breakthroughs are constantly increasing.

In natural language processing, for example, OpenAI has recently introduced GPT-3. This is a model coming from approximately 175 billion parameters, not far off the number of connections inside the human brain. 

We’ve also seen a massive improvement in Reinforcement Learning, a method to train deep learning algorithms through trials and errors, that now helps teach a vehicle to drive. The way this method works is that in the first iteration you just drive until you hit something and then the AI learns from this experience and corrects itself. After loads of different trials and errors you would eventually reach perfection and fully autonomous driving!

The amount of applications people are building with deep learning technology is enormous – it’s as though a human being is being generated! We’re seeing some very nice applications in recent years. As well as being able to generate text, diagnose medically, there are massive advancements in the automotive industry.

At Brodmann17, we are building on those advancements, as well as developing our own algorithms and methods to bring the state of artificial intelligence to the automotive industry. And this is what Brodmann17 is working on.

To learn more on deep learning, listen to the whole of Amir’s masterclass at wisdom and insights from the world’s best.

Please leave your comments below as we’d be interested to know your thoughts on deep learning and how you are applying the techniques.

Dr. Amir Alush

Dr Amir Alush is an enthusiast for algorithms coding and design. He has many years experience in designing and implementing algorithms in the fields of Computer Vision, Machine Learning and Deep Learning working on big-data. His research for his Phd focused on Machine Learning algorithms and discrete optimization problems with applications to Computer Vision. His own personal breakthroughs in deep learning culminated in him co-founding Brodmann17 to bring uncompromising AI to the edge and everyday applications.