Viewing profile — kotach

It would be innovative if they had some super fast on-line optimization of delivery routes. Optimized routes would allow them to chain several pickups from restaurants with a chain…

Vowpal Wabbit is IO limited. Meaning that there's no way it is slower than anything else on a single machine. On multiple machines it glides faster than light. So, the benchmark is…

Checkout Dagger [2], SEARN [3] and LOLS [1] (LOLS is available in vowpal wabbit search capabilities). A lot of interesting stuff on mimicking optimal policies, local optimality, jo…

If it works for chess, it'll work for Go. Chess has lots of games that you can learn from, Komodo wins any grandmaster or draws. The problem with Go was lack of evaluation function…

LSTM would converge even faster. A K-level breadth first search mimicking the optimal policy and a simple learning to search algorithm with a cost sensitive binary linear classifie…

chess grandmaster can easily be beat by a smartphone app (komodo), is smartphone using more energy than a human brain? the problem with Go is that there's little data and the game …

Studies on twins, especially those that observe separated ones, pretty much show how much bodies behave in a deterministic way, from diseases, to relationships, names, jobs, wishes…

Now, train the network jointly over the game sequence. Or even better, when given a chance to take action rollout on each action and learn jointly on that rest of gameplay. Reinfor…

Given Langford's locality vs globality argument this also gets quite obvious for the 4th game mistake and overconfidence that AlphaGo had. The rate of growth of the compounding err…

Yes, it is true. In the case of Super Mario he does the learning by simulating level-K BFS from positions that resulted in errors (unseen states) and thus minimizes the regret for …

That's not really a problem. Given a large enough dataset you want to generalize from it - there are always states not present in the dataset - the whole point is now to extract fe…

They train using trajectories but train them to guess the trajectory locally, not globally. Discounted long-term rewards are just a hack, they aren't joint learning. The concept of…

Yes, the "label bias" is more of a structured learning / joint learning term that is present in natural language processing. But reinforcement learning suffers only if you do the l…

What you are talking about here is called "label bias". [2] It is present only if training is done badly. When you have a game of Go, or Super Mario level. You don't want to make y…

I believe the whole point of pretraining on reference policies, which a collection of "optimally" played human games is, is just avoidance of bad local optimum. It can be a case th…

The questions you pose require solving the game, at least (ii). https://en.wikipedia.org/wiki/Solved_game

7.5-point komi variant played by AlphaGo and Lee has a win or lose outcome. There's no draw. But yes, a more formal definition of global optimality does not include victory as a ne…

AlphaGo is approximating global optimality by finding local optimality. Local optimality is already computationally very hard, but it is exactly what AlphaGo is doing. The rollouts…

Chess can be played godlike on a smartphone. Result of years of refining algorithms. Same could probably be accomplished with Go.

AlphaGo is certainly controllable.

I'd say I could do without concepts, modules and coroutines but ranges ! Ranges were so nice and would finally allow for easier stream handling.

AlphaGo has a learned evaluation function for each move. Evaluation function exists but it is not as simple as it can be for chess.

This comment is on-topic. Everything else is off-topic. Improve that precision!

Yeah, that's exactly the food I was thinking of. Cheeseburger + chocolate milk shake. Quinoa with kale. What would you say is more healthy? The comparison is idiotic. If a person s…

All food is healthy. Diets can be unhealthy. It really is interesting that the whole science is concentrating on a single ingredient.