4 ms·

> That is crazy fast training speed. Is it? If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadru

by retrac 5y ago

> That is crazy fast training speed. Is it? If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadrupedal robot, based on real-time feedback to its outputs? It is a simpler problem, but drone flight software based on learning, can re-learn how to fly a quadcopter after something like losing a propeller and a sudden shift in weight distribution, in a few seconds. A robot could be built that learns how to adapt to something like a limb being lost or added or half its weight shifted to the other end. It could probably learn an approximate optimal to move under those conditions in just seconds, as well. Though, I suspect that foal might actually be nearly as adaptable. Adult horses, or humans... not so much. We have many overlapping models of how to move, probably. I would take only seconds to adapt to half my weight being added on my shoulders, in the same ballpark as our best robotic systems. But if my left leg grew four inches it would take me a lot longer than the robot to learn and internalize the best way to move again. I could barely keep up with that when it was just a few inches a year when I was a kid.

teruakohatu 5y ago
> Is it? If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadrupedal robot Seems easy but why don't we have self driving cars by now? Walking it not just a mechanical task. Your quadcopter is not learning to interpret raw electrical signals from an IMS and raw electrical signals from a CMOS sensor. The training is also probably done in a simulator, where if it breaks, it is reset and training continues. A horse is learning to interpret inner ear signals and signals from the optical nerve AND the using those to build a spatial map of the environment. Only then it is actually trying to walk. A 2 hour old baby horse can not only run on not-flat terrain but is spatially and environmentally aware. It has taken years for Boston Dynamics to develop a platform that is equivalent to a pack horse and even then probably can't respond independently to never-seen-before threats or challenges.
- smegger001 5y ago
  Because the average first time driver is primed with approximately 15 years of training data from observation of not just driving deriving a complex physics model, object detection and identification system, spacial reasoning, natural language comprehensions, theory of mind able to predict behavior of other drivers based on subtle cues, and teen drivers are still awful at it as a rule with much higher accident rate. Trying to build a neural net or set of neural nets that can do all of those jobs is a daunting task and no one has the time to train the artificial neural network for 15 years. that's why we don't have safe self driving cares yet. as for the horse, well 50 millions of years of evolution on constantly refined genetic algorithm is hard to beat and that was on top of the evolution of the eohippus as a starting point
- retrac 5y ago
  We do have self-driving cars we just don't have any we think are good enough, well-tested, and adaptable to mixed-use driving and our various regulatory and safety and cultural needs. You can't get regulatory approval for a car that learns to drive by guided trial and error in the wild -- which is the most direct way of training such a thing. Feedback on what you shouldn't drive over also isn't as immediate and unambiguous, compared to whether gravity will topple it over.
lelanthran 5y ago
> If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadrupedal robot, based on real-time feedback to its outputs? It is a simpler problem, but drone flight software based on learning, can re-learn how to fly a quadcopter after something like losing a propeller and a sudden shift in weight distribution, in a few seconds. Firstly, the drone software is not re-learning anything; when a baby walks unaided for the first time and you place a rattle in their hands, do you really refer to that as "re-learning walking"? Secondly, forget about computational requirements and just look at how many iterations have to be run to reach success; mammals (and humans specifically) can "learn" to recognise an entire class of something just from one or two images of that something. Learning something else doesn't "forget" the old thing. A NN trained to recognise faces in images can't, to me knowledge, be trained to also balance a bi-pedal robot, or ride a bicycle. IOW, what I see currently is that NNs are very one-dimensional - the net that becomes optimised for one task becomes unoptimised for that task if you subsequently add new tasks.

by retrac 5y ago

teruakohatu 5y ago

> Is it? If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadrupedal robot Seems easy but why don't we have self driving cars by now? Walking it not just a mechanical task. Your quadcopter is not learning to interpret raw electrical signals from an IMS and raw electrical signals from a CMOS sensor. The training is also probably done in a simulator, where if it breaks, it is reset and training continues. A horse is learning to interpret inner ear signals and signals from the optical nerve AND the using those to build a spatial map of the environment. Only then it is actually trying to walk. A 2 hour old baby horse can not only run on not-flat terrain but is spatially and environmentally aware. It has taken years for Boston Dynamics to develop a platform that is equivalent to a pack horse and even then probably can't respond independently to never-seen-before threats or challenges.

smegger001 5y ago
Because the average first time driver is primed with approximately 15 years of training data from observation of not just driving deriving a complex physics model, object detection and identification system, spacial reasoning, natural language comprehensions, theory of mind able to predict behavior of other drivers based on subtle cues, and teen drivers are still awful at it as a rule with much higher accident rate. Trying to build a neural net or set of neural nets that can do all of those jobs is a daunting task and no one has the time to train the artificial neural network for 15 years. that's why we don't have safe self driving cares yet. as for the horse, well 50 millions of years of evolution on constantly refined genetic algorithm is hard to beat and that was on top of the evolution of the eohippus as a starting point
retrac 5y ago
We do have self-driving cars we just don't have any we think are good enough, well-tested, and adaptable to mixed-use driving and our various regulatory and safety and cultural needs. You can't get regulatory approval for a car that learns to drive by guided trial and error in the wild -- which is the most direct way of training such a thing. Feedback on what you shouldn't drive over also isn't as immediate and unambiguous, compared to whether gravity will topple it over.

lelanthran 5y ago

> If we ignore compute requirements, how long would a good algorithm take to learn something like balancing a quadrupedal robot, based on real-time feedback to its outputs? It is a simpler problem, but drone flight software based on learning, can re-learn how to fly a quadcopter after something like losing a propeller and a sudden shift in weight distribution, in a few seconds. Firstly, the drone software is not re-learning anything; when a baby walks unaided for the first time and you place a rattle in their hands, do you really refer to that as "re-learning walking"? Secondly, forget about computational requirements and just look at how many iterations have to be run to reach success; mammals (and humans specifically) can "learn" to recognise an entire class of something just from one or two images of that something. Learning something else doesn't "forget" the old thing. A NN trained to recognise faces in images can't, to me knowledge, be trained to also balance a bi-pedal robot, or ride a bicycle. IOW, what I see currently is that NNs are very one-dimensional - the net that becomes optimised for one task becomes unoptimised for that task if you subsequently add new tasks.