boltzmannbrain

Would be interested to see the results on a benchmark dataset for online anomaly detection, comparing to those approaches used in practice: https://github.com/numenta/NAB#the-numenta-anomaly-benchmark...

Max-product BP inference propagates local uncertainties in the model to arrive at a globally coherent solution. An example of where this is particularly useful is resolving border-ownership amongst neurons representing a common contour.

For the probabilistic graphical model and belief prop perspective, check out Friston et al. (2017) "The graphical brain: Belief propagation and active inference": https://www.mitpressjournals.org/doi/pdf/10.1162/netn_a_0001...

For the neural corollaries of predictive coding, check out Shipp (2016) "Neural Elements for Predictive Coding": https://www.frontiersin.org/articles/10.3389/fpsyg.2016.0179...

For a state-of-art CV framework that fits with the free energy principle, check out the Recursive Cortical Network from George et al. (2017) "A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs ": http://science.sciencemag.org/content/358/6368/eaag2612

Here you go: https://speed.python.org/

Been using boost for this (what else is new?):

https://theboostcpplibraries.com/boost.optional

https://stackoverflow.com/questions/22227839/how-to-use-boos...

Thanks that's good to know.

Haha I knew "legacy" would ruffle some feathers, by which I mean Stan was pretty much the first application-ready PPS on the block -- robust toolbox of methods, actively developed/supported.

The last slide of this presentation on SMC inference in PPLs is a nice view of the PPS landscape (not to mention the whole deck is a great intro by Lawrence Murray): http://www.it.uu.se/research/systems_and_control/education/2...

Do you know of any details on how/why probabilistic programming is used over traditional methods? And why Stan? My assumption is b/c it's a legacy framework.

Thanks. I believe he's with Oxford (although may have multiple appointments), and that video is from Microsoft Research.

Vicarious AI | https://www.vicarious.com/careers/

In addition to AI researchers, we're looking for the following (fulltime onsite in San Francisco area):

- Senior SW Engineer -- robotics and computer vision

- Senior SW Engineer -- data and application services

- Graphics Developer / Visualization Scientist

- IT Engineer

Some of the perks: private chef for all meals, unlimited vacation, company phone plan, massages, free week getaway to Hawaii every year for you and your family.

Please apply directly (referencing this post and saying Alex sent you).

Getting the discussion going with an example abstract...

Nowadays AI tools are incredibly prevalent and straightforward to integrate into projects and products. Yet the research process behind the underlying models and algorithms is a different animal. Built from theory and experimentation, research code is meant to be thrown away, and is often developed by programmers from academia. Building research into product-quality code is no small feat. Adding robots to the mix brings on another set of challenges -- integration with manufacturer's SW, hardware validation tests, succumbing to the ROS ecosystem. I'll talk about challenges and lessons coordinating teams of researchers and our workflows, with examples from our transition to a monolithic repo and development of product-quality robotics code.

Any plans on expanding to match-making model?

Neural Information Processing Systems (NIPS) is a top tier AI research conference that attracts people from all over the world, including the leading minds across academia and industry. The main utility many of us get out of it is the in-person interaction with the community -- research ideas are brainstormed, papers are shared and argued, collaborations are developed. Following Trump's immigration policies last year, there was discussion of moving NIPS '17 from Long Beach, CA to Cape Town, South Africa. This year the conference is returning to it's home Montreal; not to mention a lot of AI initiatives are making their way north of the border, unsurprisingly.

I recommend the tl;dr version from Kate Gregory at CppCon 2017, “10 Core Guidelines You Need to Start Using Now”: https://www.youtube.com/watch?v=XkDEzfpdcSg

Throwback to the milestone prizes we won in 2015: https://twitter.com/theAlexLavin/status/560103792805306369

Seconded. I worked with Astrobotic (https://www.astrobotic.com) while at CMU, and many of us have the GLXP to thank for becoming better engineers and leaders.

tl;dr Otto was almost successful in making a high-end smart door lock, but despite incredible engineering and design, bad luck (read: negligence) through an acquisition deal led Otto to run out of money.

Ali Rahimi @ ~57min

I really hope this works as promised. I'm curious if they'll add support for boost (www.boost.org) though.

I find this quite interesting, as most tech companies in the Bay Area are more of a "jungle gym" structure than a "ladder". Here they've done a good job of defining what a career in this environment looks like, setting expectations and benchmarks for career growth.

1. Have issues w/ Slack

2. Check HN to see if Slack is down, find this post

3. Attempt to slack this post to team

4. Facepalm

5. Remove headphones and shout "Slack is down!"

70 page supplementary material: http://science.sciencemag.org/content/sci/suppl/2017/10/25/s...

Reference code: https://github.com/vicariousinc/science_rcn

Paper abstract highlights the model's data efficiency several times:

Learning from few examples and generalizing to dramatically different situations are capabilities of human visual intelligence that are yet to be matched by leading machine learning models. By drawing inspiration from systems neuroscience, we introduce a probabilistic generative model for vision in which message-passing based inference handles recognition, segmentation and reasoning in a unified way. The model demonstrates excellent generalization and occlusion-reasoning capabilities, and outperforms deep neural networks on a challenging scene text recognition benchmark while being 300-fold more data efficient. In addition, the model fundamentally breaks the defense of modern text-based CAPTCHAs by generatively segmenting characters without CAPTCHA-specific heuristics. Our model emphasizes aspects like data efficiency and compositionality that may be important in the path toward general artificial intelligence.

A main difference though is Apple saw illegal music downloading as a (read: the) lifeblood of the iPod and didn't discourage it -- the iPod would not have been a success without free music. FB however is attempting to thwart people from using its platform for political agenda (or at least they make it seem), while it's not necessarily illegal to post propaganda content.

The scene today is reminiscent of Apple and illegal music downloads circa 2007:

> iPods were transactional products that could be filled with legal music, pirated music, podcasts, music of one’s own creation — it didn’t matter to anyone but the iPod’s owner. Apple sold a product that benefited from openness into a relatively lawless market and reaped the rewards [1].

Social media is largely a lawless market, and FB (amongst others) is reaping the rewards. In the case of the iPod, government regulation was the bandaid, and the cure was a new incumbent: streaming music services. Importantly, streaming music was only afforded by advances in mobile/wifi tech. So how is this like fake news in social media -- what are the analogies here? Again we’ll see government regulation with minimal effect, as noted in this WSJ article. And should there be a solution to FB’s propaganda problems, my bet is it’ll come from a service that alters the means by which we share (social) experiences online, just as the ability to stream music replaced the need to download it. What's more, the solution will likely come from outside the walls of FB.

FB will tread on though, just like Apple, by embracing the novel tech; the iPod was supplanted by the iPhone, leveraging external streaming music services like Spotify and Pandora. So it's not exactly "who will rein in FB?" but rather "what". What technology will give us new means by which we interact with social media? There are dozens of ideas, but the billion dollar one will be that which eradicates the spread of propaganda, just as streaming services eradicated the dependence on pirated music a decade ago.

[1] https://stratechery.com/2017/apple-and-the-oak-tree/

Also out this week in Science is George et al. "A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs": https://www.vicarious.com/2017/10/26/common-sense-cortex-and...

No, quite different approaches and goals.

And more on research-level techniques:

Ahmad et al. "Unsupervised real-time anomaly detection for streaming data": http://www.sciencedirect.com/science/article/pii/S0925231217...

Anomaly detection benchmark (dataset and evaluation): https://github.com/numenta/NAB

meta, cool

Coding: Talk Python to Me, Software Engineering Daily

AI: Talking Machines

Misc: Waking Up with Sam Harris, a16z

That'd be great. How can we stay in the loop to hear about Python and other feature releases?