Re: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
Ben, you haven't given us an update on how things are going with the Novamente A.I. engine lately. Is this because progress has been slow and there is nothing much to report, or you don't want to get peoples hopes up while you are still so far from being done, or that you want to surprise us one day with, "Hey guys, guess what? The Singularity has arrived!" To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] Update on Novamente progress
Hi Mike, About Novamente project progress... The reason I haven't given progress updates to this list lately is that I've been even more insanely busy than usual, due to a combination of AI work and (Novamente-related) business work and personal-life developments. So recreational emailing has fallen by the wayside of late. Now that Christmas vacation has come, I have a little time send emails! [Although I'm going on vacation for 4 days starting tomorrow, so I'll be offline for a little while] Progress on Novamente hasactually stepped up considerably as of September, when afew new people were brought into the project (in connection with a commercial application of Novamente),some of them working on language processing and one working on more fundamental AI stuff (representation of complex knowledge; and evolutionary learning). However, our recent progress has been of a technical nature, it hasn't yet yielded big milestones that are exciting to report the world at large. Regarding language processing, we have a system that "reads" English text and outputs semantic relationships contained in the text into Novamente. This is one way to fill the system's mind up with information, though from an AGI perspective it must be considered a complement to experiential interactive learning, rather than as the sole means of providing the system with knowledge. However, this "relationship extraction" software is far from complete; there's probably 6 months of work left on the "syntax" side, which will proceed in parallel with work on the semantic interpretation of the extracted relationships. Regarding reasoning, we've made a lot of progress on probabilistic inference, in the context of experimenting with inference on biological data (quantitative experimental data and data from relational DB's), and (more recently) on linguistic knowledge. This has involved a lot of technicalmath work on my part, working out various details in the inference system. The results here are very interesting, although there's a lot more testing and tweaking required, particularly regarding the control of inference. We've got the code in place for our own generalization of combinatory logic, which is the scheme we're using to represent complex knowledge in Novamente (the knowledge that would be represented using variables and quantifiers in a traditional logic-based system). And we've generalized the Bayesian Optimization Algorithm (an extension of genetic algorithms based on probability theory, invented by Martin Pelikan and David Goldberg) to learn complex combinatory logic expressions. Experimentation with this is ongoing actively, and during the first half of 2004 this code will be integrated with the probabilistic reasoning code. I spent a lot of time last month working out the nasty software-design details of integrating inferential processing with combinatory logic; that design will be implemented in January. I redesigned our previous "attention allocation" subsystem, which used to use neural net based ideas, to use a different approach based on probabilistic inference, thus simplifying the system and (I hope) inducing more emergence among components. This hasn't been implemented yet though. Yes, we STILL are not at the phase where we've hooked up Novamente to a "simulated body" in a simulated world and started teaching it experientially. We're very much looking forward to that day! As of this point, I can at least say there are no major components that are unengineered ... the coding of the generalized-combinatory-logic framework was the biggest beast AI-wise. However, there's a lot of integration, testing and tuning work ahead, and some moderate-sized beasts remain uncoded (like probabilistic attention allocation, and probabilistic logical unification). Also, there is some nitty-gritty work as yet undone, such as extending the Novamente core to run on multiple machines as a distributed system (the design was made to support extension to distributed processing, but the work hasn't been done yet). And, we've ordered our first 64-bit machine, and in early 2004 will undertake the task of porting the core to 64-bit Linux on 64-bit hardware Depending on how much attention we can give to AGI as opposed to commercial Novamente applications, we could get to the "teaching the baby" phase in mid or late 2004, or it might not be till 2005. Next, some comments on funding. Our strategy of funding AGI work via commercial applications of the in-progress AI system is working, in the sense that we're making progress implementing and testing the AI system, while getting paid for it at a reasonable level. Some of the commercial applications are also very interesting in themselves; for instance, in our bioinformatics work we've made some real progress in creating original diagnostic tools for a couple diseases (publications on this will come out in early 2004;
RE: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
Brad, Hmmm... yeah, the problem you describe is actually an implementation issue, which is irrelevant to whether one does synchronoous or asynchronous updating. It's easy to use a software design where, when a neuron sends activation to another neuron, a check is done as to whether the target neuron is over threshold. If it's over threshold, then it's put on the ready to fire queue. Rather than iterating through all neurons in each cycle, one simply iterates through those neurons on the ready-to-fire queue. Of course, one can use this approach with either synchronous or asynchronous updating. We used this design pattern in Webmind, which had a neural net aspect to its design; Novamente is a bit different, so such a strategy isn't relevant. -- Ben G While I haven't read any of the documents in question, I'd like to expound a bit here. While you are certainly correct, I think Pei was referring to the wasted computational power of updating synapses that are inactive and have no chance of being activated in the near future. In our current Neumann architectures, memory is much cheaper than CPU cycles, which is not the case in the brain. So while the brain opts for minimal neurons, and keeps most of them active in any given situation, a silicon NN might have factors of 10 more neurons, but use very sparse encoding and a well optimized update algorithm. This setup would emphasize only spending CPU time updating neurons that have a chance of being active. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
Guess I'm too used to more biophysical models in which that approach won't work. In the models I've used (which I understand aren't relevant to your approach) you can't afford to ignore a neuron or its synapses because they are under threshold. Interesting dynamics are occurring even when the neuron isn't firing. You could ignore some neurons that are at rest and hadn't received any direct or modulatory input for some time, but then you'd need some fancy optimizations to ensure you're not missing anything. But in the situation you're referring to with a more abstract (and therefore more useful to AGI) implementation, these details are irrelevant. I just wanted to chime in and ramble a bit :) Very glad to hear things are going well with Novamente. Hope the holidays treat all of you well. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
Yep, you're right of course. The trick I described is workable only for simplified formal NN models, and for formal-NN-like systems such as Webmind. It doesn't work for neural nets that more closely simulate physiology, and it also isn't relevant to systems like Novamente that are less NN-like ben -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Brad Wyble Sent: Wednesday, December 24, 2003 11:59 AM To: [EMAIL PROTECTED] Subject: RE: [agi] The emergence of probabilistic inference from hebbian learning in neural nets Guess I'm too used to more biophysical models in which that approach won't work. In the models I've used (which I understand aren't relevant to your approach) you can't afford to ignore a neuron or its synapses because they are under threshold. Interesting dynamics are occurring even when the neuron isn't firing. You could ignore some neurons that are at rest and hadn't received any direct or modulatory input for some time, but then you'd need some fancy optimizations to ensure you're not missing anything. But in the situation you're referring to with a more abstract (and therefore more useful to AGI) implementation, these details are irrelevant. I just wanted to chime in and ramble a bit :) Very glad to hear things are going well with Novamente. Hope the holidays treat all of you well. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
