@enumaris coincidentally just ran across this kaggle.com/c/dog-breed-identification

with so little data, it seems the only way to do that one would be transfer learning

maybe even look into one-shot learning

also, it seems that the two problems are the same "in randomized reduction". But it seems strange; since both of which are not complete problems in this class and we knew that If P != NP, then there must be problems in this class in NP-intermediate. I also conjecture that if there were problems in NP-intermediate, then there must be a complete problems!

reddit - compsci - Thoughts on Jarek Duda's paper about P?=NP and graph isomorphism problem?

reddit - programming - Jarek Duda (known from ANS coding and being screwed by Google) shows what he claims is close to polynomial algorithm for graph isomorphism problem

Thanks, anything interesting there? I mostly see basics and Google :/

I have just added answer to the first one, but don't want to touch the second

I don't know. I guess it is one of the typical reddit threads. That sort of thread was part of the motivation behind the "amateur reviews" chatroom.

I have question regarding "computation" and "algorithm". you can notice that in literature we have ("Approximation Computing" and "Approximation Algorithm") or ("Parallel Computing" and "Parallel Algorithm") or ("exact computing=Church-Turing Thesis" and "exact algorithm= simply called 'algorithm'") or ("Distributed Computing" or "Distributed Algorithm") etc.

Now, when you design an algorithm, you need to know the model of computation whether it is quantum computation model, exact computation model, etc. I want to know the exact work of people who work on "computation". Does they work "physically" how to design a model of computation say for quantum model or for randomized model? it seems for me, designing computation model is kind of people who are working on computer architecture. Is that correct! I just want to make sure!

Explicit Multi-Threading (XMT): A PRAM-On-Chip Vision

Algorithmic Approach to Designing an Easy-To-Program System: Can It Lead to a HW-Enhanced Programmer’s Workflow Add-On?

Uzi Vishkin argued his case from an economic perspective:

> Alas, the software spiral is now broken: (a) nobody is building hardware that provides improved performance on the old serial software base; (b) there is no broad parallel computing application software base for which hardware vendors are committed to improve performance; and (c) no agreed-upon architecture currently allows application programmers to build such software base for the future

@YOUSEFY There are nice theoretical models for parallel and serial computing. Sadly, only the serial model has really been picked up by current hardware. In practice, MPI provides a very reasonable model for distributed computation. The current parallel hardware is still quite CCNUMA, which is a bad mix between parallel and distributed.

But I guess your question is more motivated by quantum computing, which seems to be simulataneously reversible, parallel, and randomized. So you wonder whether it would not make sense to first understand each of those paradigms on its own, i.e. a randomized hardware architecture, a reversible hardware architecture, and a parallel hardware architecture.

And you are right, quantum computing uses the circuit model, a prime example of a parallel computer architecture model. Quantum algorithms use randomness and are compared against classical randomized algorithms. Only the reversible part seems to be mostly ignored, but even this is not fully true. There are steps to uncompute results in auxiliary memory, which occur in both QC and RC, even if they are often glossed over.