interesting ... out of curiosity. what type problems might this approach solve that would not be solved with current methods like those in sympy? there's bound to be examples but i couldn't think of any on the spot.
-
-
-
I doubt there’s any problem Mathy will solve that sympy wouldn’t, but the steps Mathy outputs along the way are important. Imagine you’re working on monomials for homework in a math class. You’re stuck and need to see other examples worked help your understanding:
- Još 7 drugih odgovora
Novi razgovor -
-
-
Super cool! Stupid question: where does the training data/cost function for this come from? This seems like a situation where one could create infinite synthetic data (for example given a simplified expression, generate N equal expressions which are more complicated)
-
Thanks!
The training data comes from observations of the environment the agent is working in. You define a function for generating problem texts, and another for checking for terminal states
Problems: https://mathy.ai/envs/overview/#new-problems …
Transitions: https://mathy.ai/envs/overview/#custom-win-conditions … - Još 2 druga odgovora
Novi razgovor -
-
-
I call this cool, not scary! :)
-
Thanks! It's scary because I'm not a Math or Machine Learning expert, so I may have missed a bunch of silly, obvious things in my implementation.

- Još 6 drugih odgovora
Novi razgovor -
-
-
This is really cool! You wrote your own CAS? How did you find that? What's your goal with this? What about RL algos, what are limitations & failure rates? What are frustrating limitations you wish there was a solution for? As someone doing something that intersects, curious!
-
Thanks! The original Tokenizer/Parser/Expressions came from Bob Brown
@Microsoft https://mathy.ai/about/#silverlight-graphing-calculator … Probably the biggest limitations are my lack of expertise in reinforcement learning. I haven't studied agent perf formally, but here's an evaluation run I did a while ago:pic.twitter.com/zdtKfNMBlD
Kraj razgovora
Novi razgovor -
-
-
phenomenal work!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
@JusteRaimbault pour toi ? -
Would be fun to put such agents in some kind of
#ArtificialLife coevolution/biogeography algorithm: do different communities construct concurrent algebra; is there a global optimum at an intermediate modularity, etc. ?
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
