It was an event with really engaged speakers, moderators, and
Therefore, some things remained stuck in my mind until the day after, so they are probably worth sharing. (For a broad coverage, see https://twitter.com/hashtag/ai2015).
As Dr Rand Hindi from Snips puts it, the overwhelming complexity arising from the exponentially growing number of connected devices creates friction that breaks grounds for AI. Hard-coded or manual attempts to solving these issues (friction, interruptions) just do not work anymore. In other words, it seems like the good old “when the going gets tough, the tough
The keywords here are context-awareness and privacy by design. In my opinion, privacy by design as a core principle is the right way forward to get broad traction among consumers. It puts current predominant “internet” business models at stake, but luckily other monetization (or better compensation) models are there. The biggest move towards privacy (and some ubiquitous tech) in this context, is probably GSMA’s Mobile Connect. Let’s see what 2015 brings in this respect.
Dr Gabriel Brostow’s approach of short-cutting the learning phase, by basing on previously gathered knowledge, breaks ground for broader and more spontaneous use of machine learning.
#realtime #localisation #orientation
The real-time mapping capabilities demonstrated by Dr Stefan Leuteneger were promising. Highly accurate and compact enough to be incorporated into robots, devices and whatsoever.
#routing # objectives #mapping
The happy maps are a good thing and a reminder of what counts. There is and always will be a conscious decision on what is important – speed, travel time, safety, insurance incentives, pleasure and benefits. Thanks to Dr Daniele Quercia for the entertaining presentation and discussing self-determination and manipulation.
#ai #saas #opensource
Seldon surprised me with their efforts to open up AI. Pr
#conversational #voicerecognition #neuroscience #efficiency
Now that’s a tag list! It’s my melting pot for all topics towards smartness of systems. The first one is bringing conversational speech recognition to life (“no means no, so don’t ask twice” – as presented by Dr Blaise Thomson). The second is using fractions of information to conclude the full speech (as discussed on the example of following a specific speaker in a noisy environment by Dr Dan Goodman). They are showing a way from fully-specified systems and data sets (which are so nice and neat and so…unrealistic) towards real-world scenarios. It is making systems more resilient to “unknown Unknowns” – and will bring relief from exception catching and handling.
#deepmind #google #selflearning #businesscase
I’ve never been very good at Arcade games. Hence, I really enjoyed the demonstration of the learning curve of Deepmind’s General Deep Learning approach when applied to a test-bed of some 100+ Atari games, vividly presented by Mustafa Suleyman. It was amazing how well the system performed even without supervised training, just being introduced to the controls (= 4-way joystick) and the objective (= the score). With all due respect and scepticism, it is astonishing that they managed to replace some 60 hard-coded production systems with an unsupervised system based on machine-learning. The systems I recount were image recognition, image/pattern/structure matching and such.
I admit that the discussion of “the potential end of mankind led by super-machines” is not my cup of tea. IMHO, we should keep a good eye on the social impact of modern technology in all it’s variations – and should never forget: >tech is tool<.
Please feel free to add and comment. Definitively, I hope the guys from Playfair Capital keep this event up – well done!
(Please note: I am not involved in or otherwise related to the above products/services/companies; any mentioning or not mentioning is purely based on my very own attentional filter.)