Profile cover photo
Profile photo
Falcon Dai
Observe, don't assume.
Observe, don't assume.

Falcon's posts

Post has shared content
i like the dreams of machines idea. more precisely, the idea that dreams ~ fixed points in the state space of some autoencoders 
Artificial Neural Networks have spurred remarkable recent progress in image classification and speech recognition. But even though these are very useful tools based on well-known mathematical methods, we actually understand surprisingly little of why certain models work and others don’t.

Over on the Google Research blog, we take a look at some simple techniques for peeking inside these networks, yielding a qualitative sense of the level of abstraction that particular layers of neural networks have achieved in their understanding of images. This helps us visualize how neural networks are able to carry out difficult classification tasks, improve network architecture, and check what the network has learned during training. 

It also makes us wonder whether neural networks could become a tool for artists—a new way to remix visual concepts—or perhaps even shed a little light on the roots of the creative process in general.

Post has attachment
Don't miss a game. A chrome extension that notifies you World Cup matches.

Post has attachment
Quick & easy way to mount Google Glass onto eyeglasses
I recently got my Google Glass. I was very excited about the product and the future of head mount displays in general but a very practical problem arose as soon as I tried to setup the device: I wear prescription lenses and without them I cannot see anythin...

hi fellow explorers. i am a developer/researcher at UChicago. i am looking forward to working on Glassware dev (or deeper hacking).

Post has shared content
DeepMind showcased an awesome demo of AI agent palying arcade games at human expert level at NIPS 2013.

Post has shared content
Closer to AI
Many of you have already seen this Nature article about deep learning.
It's a nice and largely accurate article, with a few inevitable mistakes. 

Since I have taken the bad habit of commenting on media coverage of AI, I feel compelled to do so here. I must disclose that I was interviewed by the author for this article Nicola Jones. She tried to contact me for some fact checking after she wrote the article, but I was traveling and could not respond. 

I feel bad for +Yoshua Bengio that Nicola chose to not mention his role, given the big impact he has had on deep learning. I also feel bad that she chose not to mention the crucial role of the Canadian Institute for Advanced Research (CIFAR, a private foundation). For the last 10 years, CIFAR has been funding a program called Neural Computation and Adaptive Perception. CIFAR-NCAP essentially bankrolled the initial phase of the "deep learning conspiracy" fomented by Geoff Hinton, Yoshua and me about 10 years ago, whose purpose was to revive the interest of the machine learning and AI community in the problem of learning representations. NCAP partly funded the research of a number of people interested in (deep) learning and perception (artificial and natural), and sponsored workshops and summer schools that were key to bringing deep learning to life as a community.

The press keeps being fascinated by the "Google cat detector". But this work has had a relatively small impact on the field, and essentially no impact on the practical applications of deep learning. It was a nice experiment in large-scale unsupervised learning, but the error rate it yields for object recognition is not competitive, and Google doesn't use the method in practice. 

The press also likes to mention the fact that this system was trained on "1,000 computers", without mentioning that the practical applications of deep learning can actually be trained on a single sub-$1000 GPU card designed for gamers.

I am quoted in the article as saying that "AI has gone from failure to failure, with bits of progress". I don't remember being that harsh with our field! What I said is that with every new wave of AI came unreasonable expectations caused by hype (sometimes by the press, sometimes by people trying to profit financially, sometimes by overly optimistic researchers). When expectations are not met, there is an inevitable backlash, and funding agencies, investors, and potential customers turn away. But these waves are never complete failures: they leave behind them many useful new tools in the toolbox that find their way into applications. Perhaps these applications are not as flashy as the hype made us believe, but they are useful.

Is there anyone attending NIPS still wants a roommate or could accept one? 

Post has attachment
fixing the infamous generic IMEI (004999010640000) for Galaxy Nexus

Post has attachment
Fixing Galaxy Nexus's IMEI Number
I have been using Galaxy Nexus for quite some time and it worked fine as an everyday+development phone until its IMEI number got blocked on T-Mobile recently. If your IMEI number is  004999010640000 (you can see your IMEI number by dialing *#06#), then this...
Wait while more posts are being loaded