Incoming... Machine Learning Questions

Question

There has been strong support to officially merge Machine Learning with Cross Validated. Later this week, we will be migrating questions from the Machine Learning and making it an official part of this site.

I re-posted the Machine Learning announcement as an answer below, for your reference.

Be prepared to integrate these questions and this subject into your site. You can close questions that are blatantly off topic, but be mindful — this is a growing scope of questions looking for a best fit. We want to make every effort to accommodate the community on this site, and err towards being open-minded and make adjustments to accommodate the subject.

Answer 1

This is a re-post of the Machine Learning announcement (since it is still in private beta):

---

The Future of Machine Learning as a Stack Exchange Subject

The "Machine Learning" Stack Exchange started with a proposal in Area 51. A groups of experts propose a subject they felt were not found on the Stack Exchange Network. In the meantime, you have a Stats SE ("Cross Validated") who claim that "Machine Learning" is explicitly on topic for that site. Still, the supporters of the ML site wonders if all machine learning topics are a good fit for Cross Validated.

So what to do? We're not experts in this domain, so we let the proposal continue to see how it develops in actual practice. As it turns out, about 85-95% of the questions being asked are a perfect fit for Cross Validated. Even now, there is still widespread support to welcome the Machine Learning questions on their site.

Should Machine Learning SE be merged with CrossValidated?

So be it. Around mid week, we are going to migrate the questions from "Machine Learning" to the Cross Validated site. Some will be closed as off topic, but for the most part, they would love to see your Machine Language questions there. Make Machine Learning a strong subject on the Cross Validated site by asking your questions there.

For the smaller group of questions that do not fit on that site, support the Computer Science proposal. Bring your questions to these sites and make your presence known in the tags you care about. There is no purpose in continuing a sub-site like this when the subject is well-covered elsewhere.

A larger, more comprehensive site is a win-win to everyone.

Answer 2

For informational purposes, the list of questions that were migrated were:

• Classification with tall fat data
• Why is there an asymmetry between the training step and evaluation step?
• Possible to get a better ANN by removing some connections?
• Fast method for finding best metaparameters of SVM (that is faster than grid search)
• Creating a maximum entropy Markov model from an existing multi-input maximum entropy classifier
• What is a good method for designing an HTM network in terms of depth and breadth?
• Domain-agnostic feature engineering that retains semantic meaning?
• Probabilistic (Bayesian) vs Optimisation (Frequentist) methods in Machine Learning
• What's the correct way to test the significance of classification results
• What is the proper naming scheme for dataset parts?
• Why do naive Bayesian classifiers perform so well?
• Applying machine learning for DDoS filtering
• On what tasks does neuroevolution outperform basic application of neural networks or genetic algorithms?
• What are alternatives to VC-dimension for measuring the complexity of neural networks?
• Are there algorithms for computing "running" linear or logistic regression parameters?
• Why do we assume that the error is normally distributed?
• How to decide on the correct number of clusters?
• Does a sparse training set adversely affect an SVM?
• Problem specific regularization
• Probability and log probability in hidden Markov models
• Are machine learning techniques "approximation algorithms"?
• Is there a Project Euler-alike for machine learning?
• Dealing with datasets with a variable number of features
• Are associative networks still pertinent for data mining applications?
• What language to use for genetic programming
• How to create recommender system that integrates both collaborative filtering and content features?
• Why does Natural Language Processing not fall under Machine Learning domain?
• Modern neural networks that build their own topology
• How would you design a machine learning system to play Angry Birds?
• Are MFCCs the optimal method of representing music to a retrieval system?
• Fiducial Inference in Machine Learning
• At what n do n-grams become counterproductive?
• Classification of observation symbols in a HMM?
• Automatic feature selection for anomaly detection
• Multilayer neural networks for multivariate temporal data
• Is whitening always good?
• Dynamic recommender systems
• Classifier success rate and confidence intervals
• How can I help ensure testing data does not leak into training data?
• Decision trees and backward pruning
• Difference between dynamic programming and temporal difference learning in reinforcement learning
• Subtree replacement vs subtree raising
• Automatic data cleansing
• https://stats.stackexchange.com/q/23402
• Approaches when learning from huge datasets?
• Kernel matrix normalisation
• What is standing in the way of pulsed neural networks being used in applications?
• CCA/KCCA for more than two views
• How does a Support Vector Machine (SVM) work?
• In boosting, why are the learners "weak"?
• What is a kernel and what sets it apart from other functions
• Independence in classification
• Best bandit algorithm?
• What are probabilistic approaches to finding the right number of clusters?
• What is the Drosophila of AI now?
• Are experts harmful?
• Beyond Fisher kernels
• Dealing with very large time-series datasets
• Classification of conversations based on content
• Where do I find large face datasets?
• Probabilistic outputs from SVMs
• Online clustering
• Viable distance metric for text articles
• Continuous Multistate Ada-boost method?
• MRF MAP inference for non-submodular pairwise terms
• How to perform Gaussian process regression when function being approximated changes over time?
• PCA, LDA, CCA, and PLS
• Large performance variance when using back-propagation in neural networks for feature extraction
• Average case analysis of learning algorithms
• Dealing with mixed categorical data: e.g. Heritage Health Prize data
• Number of states and symbols in multi class Hidden Markov Model classifier