Intel organized its own version of AI day on 4th April. For those who couldn’t make it in person, the event was webcast live. Various Intel leaders spoke about the future of AI, associated challenges and how Intel was best positioned as an end to end player in the acceleration of AI.
Intel’s vision: Provide best silicon, best software and best ecosystem for accelerating AI adoption.
It was quite an informative event providing a glimpse into Intel’s AI ecosystem across hardware & software spectrum. Exciting times ahead for AI I must say! In between listening to these presenters, I managed to pick some bits and bytes that I would like to share with everyone.
- Definition of AI: the ability to reason, sense and act on its own without requiring human intervention.
- Division of labor between man and machines is challenged by machines.
- Lots of data, lots of computing & connectivity is the reason why AI is growing today.
- Connection of supercomputing to the masses, real-time feedback & real-time inference will be real AI revolution. AI will be an edge play.
- Classical machine learning requires someone to tell the features. How do we know the right features? That’s why AI has stayed in books for long.
- Promise of deep learning – I don’t need experts to tell me the features or rules. Deep learning learns new features on its own & features may change over time.
- Deep learning currently works for speech and images very well- to the extent, it beats all other algorithms.
- Less data with less supervision is our first research goal.
- Deep learning models today can’t reason. We are looking to augment deep learning models with Bayesian & Reinforcement models so it becomes easier to reason.
- Scarification & prunification for optimal deployment of a model is a very significant research area.
- AI compute cycles will grow 12 times by 2020. AI will be the largest workload in data centers.
- Intel’s Nervana Platform will deliver 100 X improvement in training time for deep learning by 2020.