The Curious Engineer

All blogs / AI / ML

Machine Learning (Theory)

hunch.net

Machine learning and learning theory research

AI / MLScience

’05’10’15’20’25
Posts
549 from 19 Jan 2005 to 21 Sep 2026
Posting
now and then (2 in the last year)
Kind
personal blog
Feed
https://hunch.net/?feed=rss2

2026 2

  1. Curbing AI with Flight Recorders, Brainwash Entrapment Tests, and 3rd Party Evaluation
  2. Agentic Coding Ascension Notes

2025 1

  1. Headroom for AI development

2023 1

  1. An AI Miracle Malcontent

2021 2

  1. ICML 2021 Invited Speakers — ML for Science
  2. ALT Highlights – An Interview with Joelle Pineau

2020 7

  1. What is the Right Response to Employer Misbehavior in Research?
  2. Experiments with the ICML 2020 Peer-Review Process
  3. HOMER: Provable Exploration in Reinforcement Learning
  4. Critical issues in digital contract tracing
  5. What is the most effective policy response to the new coronavirus pandemic?
  6. Coronavirus and Machine Learning Conferences
  7. Updates for the new decade

2019 2

  1. ICML has 3(!) Real World Reinforcement Learning Workshops
  2. Code submission should be encouraged but not compulsory

2018 8

  1. FAQ on ICML 2019 Code Submission Policy
  2. ICML 2019: Some Changes and Call for Papers
  3. Please vote
  4. A Real World Reinforcement Learning Research Program
  5. When the bubble bursts…
  6. Reinforcement Learning Platforms
  7. ICML Board and Reviewer profiles
  8. Pervasive Simulator Misuse with Reinforcement Learning

2017 7

  1. Vowpal Wabbit 8.5.0 & NIPS tutorial
  2. The Real World Interactive Learning Tutorial
  3. ICML is changing its constitution
  4. Machine Learning the Future Class
  5. Fact over Fiction
  6. The Decision Service is Hiring
  7. EWRL and NIPS 2016

2016 12

  1. Vowpal Wabbit version 8.3 and tutorial
  2. ICML 2016 videos and statistics
  3. ICML 2016 was awesome
  4. The Multiworld Testing Decision Service
  5. An ICML unworkshop
  6. The ICML 2016 Space Fight
  7. Room Sharing for ICML (and COLT, and ACL, and IJCAI)
  8. Quora session
  9. ICML registration is live
  10. AlphaGo is not the solution to AI
  11. Learning to avoid not making an AI
  12. New York Machine Learning Deadlines

2015 5

  1. Interesting things at NIPS 2015
  2. CNTK and Vowpal Wabbit tutorials at NIPS
  3. ICML 2016 in NYC and KDD Cup 2016
  4. Randomized experimentation
  5. The NIPS experiment

2014 10

  1. Vowpal Wabbit 7.8 at NIPS
  2. Allreduce (or MPI) vs. Parameter server approaches
  3. Conference on Digitial Experimentation
  4. No more MSR Silicon Valley
  5. Open Machine Learning Workshop, August 22
  6. The perfect candidate
  7. Interesting papers at ICML 2014
  8. An ICML proposal: yearly surveys
  9. The New York ML Symposium, take 2
  10. Metacademy: a package manager for knowledge

2013 16

  1. NIPS tutorials and Vowpal Wabbit 7.4
  2. Ben Taskar is gone
  3. Graduates and Postdocs
  4. No NY ML Symposium in 2013, and some good news
  5. Extreme Classification workshop at NIPS
  6. ICML 2012 videos lost
  7. Thoughts on Artificial Intelligence
  8. The Benefits of Double-Blind Review
  9. Representative Reviewing
  10. The Large Scale Learning class notes
  11. COLT and ICML registration
  12. NEML II
  13. I’m a bandit
  14. Remote large scale learning class participation
  15. NYU Large Scale Machine Learning Class
  16. Deep Learning 2012

2012 25

  1. Simons Institute Big Data Program
  2. ML Symposium and Strata/Hadoop World
  3. 7th Annual Machine Learning Symposium
  4. Vowpal Wabbit, version 7.0
  5. NYAS ML 2012 and ICML 2013
  6. Patterns for research in machine learning
  7. MUCMD and BayLearn
  8. Videolectures
  9. ICML survey and comments
  10. Normal Deviate and the UCSC Machine Learning Summer School
  11. ICML acceptance statistics
  12. ICML accepted papers and early registration
  13. Microsoft Research, New York City
  14. ICML: Behind the Scenes
  15. Both new: STOC workshops and NEML
  16. ICML author feedback is open
  17. David Waltz
  18. The Submodularity workshop and Lucca Professorship
  19. COLT/ICML Open Questions and ICML Instructions
  20. Key Scientific Challenges and the Franklin Symposium
  21. ICML+50%
  22. Berkeley Streaming Data Workshop
  23. ICML Posters and Scope
  24. Why COLT?
  25. Why ICML? and the summer conferences

2011 31

  1. Vowpal Wabbit version 6.1 & the NIPS tutorial
  2. Hadoop AllReduce and Terascale Learning
  3. Giving Thanks
  4. 2011 ML symposium and the bears
  5. ML Symposium and ICML details
  6. Monday announcements
  7. Somebody’s Eating Your Lunch
  8. KDD and MUCMD 2011
  9. Fall Machine Learning Events
  10. The Large Scale Learning Survey Tutorial
  11. Vowpal Wabbit 6.0
  12. Interesting thing at UAI 2011
  13. Interesting papers at COLT 2011
  14. Interesting Neural Network Papers at ICML 2011
  15. ICML 2011 and the future
  16. Ultra LDA
  17. Research Directions for Machine Learning and Algorithms
  18. CI Fellows, again
  19. ICML workshops due
  20. The End of the Beginning of Active Learning
  21. A paper not at Snowbird
  22. The Heritage Health Prize
  23. COLT open questions
  24. Vowpal Wabbit, v5.1
  25. KDD Cup 2011
  26. The Ideal Large Scale Learning Class
  27. Yahoo! Machine Learning grant due March 11
  28. What does Watson mean?
  29. User preferences for search engines
  30. 2011 Summer Conference Deadline Season
  31. Herman Goldstine 2011

2010 35

  1. NIPS 2010
  2. Vowpal Wabbit, version 5.0, and the second heresy
  3. Traffic Prediction Problem
  4. ICML 2011 – Call for Tutorials
  5. To Vidoelecture or not
  6. NY ML Symposium 2010
  7. Partha Niyogi has died
  8. An easy proof of the Chernoff-Hoeffding bound
  9. Machined Learnings
  10. Regretting the dead
  11. New York Area Machine Learning Events
  12. AIStats
  13. Alex Smola starts a blog
  14. Boosted Decision Trees for Deep Learning
  15. KDD 2010
  16. Rob Schapire at NYC ML Meetup
  17. The Workshop on Cores, Clusters, and Clouds
  18. ICML & COLT 2010
  19. MetaOptimize
  20. 2010 ICML discussion site
  21. The Good News on Exploration and Learning
  22. Google Predict
  23. Aggregation of estimators, sparsity in high dimension and computational feasibility
  24. What’s the difference between gambling and rewarding good prediction?
  25. CI Fellows program renewed
  26. Compassionate Reviewing
  27. COLT Treasurer is now Phil Long
  28. MLcomp: a website for objectively comparing ML algorithms
  29. A Variance only Deviation Bound
  30. The Efficient Robust Conditional Probability Estimation Problem
  31. Netflix Challenge 2 Canceled
  32. Yahoo! ML events
  33. Specializations of the Master Problem
  34. Deadline Season, 2010
  35. Sam Roweis died

2009 52

  1. Interesting things at NIPS 2009
  2. Top graduates this season
  3. Inherent Uncertainty
  4. Future Publication Models @ NIPS
  5. Vowpal Wabbit version 4.0, and a NIPS heresy
  6. AI Safety
  7. ICML 2009 Workshops (and Tutorials)
  8. The Other Online Learning
  9. NYAS ML Symposium this year.
  10. Yisong Yue on Self-improving Systems
  11. NIPS workshops
  12. ALT 2009
  13. Static vs. Dynamic multiclass prediction
  14. Machine Learning Protests at the G20
  15. Netflix finishes (and starts)
  16. Necessary and Sufficient Research
  17. New York Area Machine Learning Events
  18. Another 10-year paper in Machine Learning
  19. Centmail comments
  20. Carbon in Computer Science Research
  21. Vowpal Wabbit Open Source Project
  22. Interesting papers at KDD
  23. The Machine Learning Forum
  24. Netflix nearly done
  25. Interesting papers at UAICMOLT 2009
  26. In Active Learning, the question changes
  27. Functionally defined Nonlinear Dynamic Models
  28. Multitask Poisoning
  29. Many ways to Learn this summer
  30. 2009 ICML discussion site
  31. CI Fellows
  32. Server Update
  33. Computability in Artificial Intelligence
  34. Machine Learning to AI
  35. Wielding a New Abstraction
  36. Jonathan Chang at Slycoder
  37. Interesting Presentations at Snowbird
  38. Asymmophobia
  39. Machine Learning is too easy
  40. Parallel ML primitives
  41. Prediction Science
  42. Effective Research Funding
  43. Decision by Vetocracy
  44. KDNuggets
  45. Optimal Proxy Loss for Classification
  46. Nielsen’s talk
  47. Key Scientific Challenges
  48. An Active Learning Survey
  49. Nearly all natural problems require nonlinearity
  50. Netflix prize within epsilon
  51. Predictive Analytics World
  52. Interesting Papers at SODA 2009

2008 52

  1. Adversarial Academia
  2. Use of Learning Theory
  3. Summer Conferences
  4. A NIPS paper
  5. A Bumper Crop of Machine Learning Graduates
  6. Efficient Reinforcement Learning in MDPs
  7. Observations on Linearity for Reductions to Regression
  8. COLT CFP
  9. ICML Reviewing Criteria
  10. A Healthy COLT
  11. Rise of the Machines
  12. New York’s ML Day
  13. NIPS 2008 workshop on Kernel Learning
  14. Who is Responsible for a Bad Review?
  15. NIPS 2008 workshop on ‘Learning over Empirical Hypothesis Spaces’
  16. The SODA Program Committee
  17. How do we get weak action dependence for learning with partial observations?
  18. Fall ML Conferences
  19. Bidding Problems
  20. Mass Customized Medicine in the Future?
  21. Radford Neal starts a blog
  22. Electoralmarkets.com
  23. Compositional Machine Learning Algorithm Design
  24. Interesting papers at COLT (and a bit of UAI & workshops)
  25. Interesting papers, ICML 2008
  26. To Dual or Not
  27. More Presentation Preparation
  28. Proprietary Data in Academic Research?
  29. ICML has a comment system
  30. Reviewing Horror Stories
  31. The Minimum Sample Complexity of Importance Weighting
  32. Inappropriate Mathematics for Machine Learning
  33. Three levels of addressing the Netflix Prize
  34. Concerns about the Large Scale Learning Challenge
  35. Watchword: Supervised Learning
  36. Eliminating the Birthday Paradox for Universal Features
  37. Taking the next step
  38. The Science 2.0 article
  39. Blog compromised
  40. It Doesn’t Stop
  41. Interactive Machine Learning
  42. COLT Open Problems
  43. Spock Challenge Winners
  44. The Stats Handicap
  45. The Meaning of Confidence
  46. Complexity Illness
  47. Sufficient Computation
  48. Turing’s Club for Machine Learning
  49. Why Workshop?
  50. Datasets
  51. 2008 Summer Machine Learning Conference Schedule
  52. Research Political Issues

2007 57

  1. Vowpal Wabbit Code Release
  2. Cool and Interesting things at NIPS, take three
  3. Cool and interesting things seen at NIPS
  4. New Machine Learning mailing list
  5. Workshop Summary—Principles of Learning Problem Design
  6. Learning Track of International Planning Competition
  7. The Netflix Crack
  8. Computational Consequences of Classification
  9. MLSS 2008
  10. BellKor wins Netflix
  11. CMU wins DARPA Urban Challenge
  12. The Machine Learning Award goes to …
  13. Contextual Bandits
  14. Second Annual Reinforcement Learning Competition
  15. Online as the new adjective
  16. NIPS workshops extended to 3 days
  17. NIPS workshp: Learning Problem Design
  18. NIPS workshops are out.
  19. It’s MDL Jim, but not as we know it…(on Bayes, MDL and consistency)
  20. Optimizing Machine Learning Programs
  21. Live ML Class
  22. The Privacy Problem
  23. Choice of Metrics
  24. Exponentiated Gradient
  25. Asking questions
  26. Motivation should be the Responsibility of the Reviewer
  27. The View From China
  28. ICML Trends
  29. Idempotent-capable Predictors
  30. Watchword: Online Learning
  31. Interesting Papers at ICML 2007
  32. Machine Learning Jobs are Growing on Trees
  33. Presentation Preparation
  34. How is Compressed Sensing going to change Machine Learning ?
  35. Interesting Papers at COLT 2007
  36. Not Posting
  37. Loss Function Semantics
  38. The Missing Bound
  39. Conditional Tournaments for Multiclass to Binary
  40. COLT 2007
  41. The Coming Patent Apocalypse
  42. Videolectures.net
  43. $50K Spock Challenge
  44. What to do with an unreasonable conditional accept
  45. Contextual Scaling
  46. Alternative Machine Learning Reductions Definitions
  47. All Models of Learning have Flaws
  48. Create Your Own ICML Workshop
  49. The Forgetting
  50. 24
  51. Best Practices for Collaboration
  52. Thoughts regarding “Is machine learning different from statistics?”
  53. Parallel Machine Learning Problems
  54. The Machine Learning Department
  55. A Deep Belief Net Learning Problem
  56. 2007 Summer Machine Learning Conferences
  57. Retrospective

2006 79

  1. Interesting Papers at NIPS 2006
  2. The Spam Problem
  3. Recruitment Conferences
  4. Structural Problems in NIPS Decision Making
  5. Continuizing Solutions
  6. Explicit Randomization in Learning algorithms
  7. Context and the calculation misperception
  8. Data Linkage Problems
  9. 2006 NIPS workshops
  10. Exemplar programming
  11. David Pennock starts Oddhead
  12. Incompatibilities between classical confidence intervals and learning.
  13. Health of Conferences Wiki
  14. $1M Netflix prediction contest
  15. Programming Languages for Machine Learning Implementations
  16. Luis von Ahn is awarded a MacArthur fellowship.
  17. What is missing for online collaborative research?
  18. Incentive Compatible Reviewing
  19. How to solve an NP hard problem in quadratic time
  20. Objective and subjective interpretations of probability
  21. Learning Theory standards for NIPS 2006
  22. Report of MLSS 2006 Taipei
  23. Precision is not accuracy
  24. The Call of the Deep
  25. AOL’s data drop
  26. Two more UAI papers of interest
  27. Upcoming conference
  28. A Winner
  29. Regression vs. Classification as a Primitive
  30. Who is having visa problems reaching US conferences?
  31. New Models
  32. The Stock Prediction Machine Learning Problem
  33. MaxEnt contradicts Bayes Rule?
  34. Some recent papers
  35. Branch Prediction Competition
  36. more icml papers
  37. ICML papers
  38. Presentation of Proofs is Hard.
  39. Online convex optimization at COLT
  40. Regularization = Robustness
  41. IJCAI is out of season
  42. Explorations of Exploration
  43. Server Shift, Site Tweaks, Suggestions?
  44. What is the best regret transform reduction from multiclass to binary?
  45. NIPS paper evaluation criteria
  46. The value of the orthodox view of Boosting
  47. Big machine learning
  48. An ICML reject
  49. A conversation between Theo and Pat
  50. John Langford –> Yahoo Research, NY
  51. Conferences, Workshops, and Tutorials
  52. Rexa is live
  53. JMLR is a success
  54. Progress in Machine Translation
  55. Bounds greater than 1
  56. What is state?
  57. Mad (Neuro)science
  58. Gradients everywhere
  59. NLPers
  60. The Approximation Argument
  61. Multitask learning is Black-Boxable
  62. Online learning or online preservation of learning?
  63. Use of Notation
  64. “Structural” Learning
  65. Why do people count for learning?
  66. The Peekaboom Dataset
  67. A Fundamentalist Organization of Machine Learning
  68. Multiplication of Learned Probabilities is Dangerous
  69. Yahoo’s Learning Problems.
  70. Pittsburgh Mind Reading Competition
  71. Research Budget Changes
  72. Introspectionism as a Disease
  73. Should the Input Representation be a Vector?
  74. 1 year
  75. On Coding via Mutual Information & Bayes Nets
  76. Is Multitask Learning Black-Boxable?
  77. Benchmarks for RL
  78. Debugging Your Brain
  79. MLTV

2005 145

  1. Deadline Season
  2. Yet more nips thoughts
  3. Automated Labeling
  4. Yes , I am applying
  5. Workshops as Franchise Conferences
  6. More NIPS Papers II
  7. More NIPS Papers
  8. Some NIPS papers
  9. Machine Learning Thoughts
  10. Is the Google way the way for machine learning?
  11. Watchword: model
  12. The Webscience Future
  13. A question of quantification
  14. The Design of an Optimal Research Environment
  15. MLSS 2006
  16. The Everything Ensemble Edge
  17. Prediction Competitions
  18. The design of a computing cluster
  19. Progress in Active Learning
  20. Fallback Analysis is a Secret to Useful Algorithms
  21. Machine Learning in the News
  22. Workshop: Atomic Learning
  23. Complexity: It’s all in your head
  24. Site tweak
  25. The unrealized potential of the research lab
  26. Predictive Search is Coming
  27. We have a winner
  28. On-line learning of regular decision rules
  29. Not ICML
  30. Research in conferences
  31. Prediction Bounds as the Mathematics of Science
  32. Workshop Proposal: Atomic Learning
  33. NIPS Workshops
  34. The Predictionist Viewpoint
  35. Fast Gradient Descent
  36. “Failure” is an option
  37. Online Learning as the Mathematics of Accountability
  38. A link
  39. Site Update
  40. Science in the Government
  41. (Dis)similarities between academia and open source programmers
  42. Do you believe in induction?
  43. SVM Adaptability
  44. Why Manifold-Based Dimension Reduction Techniques?
  45. Apprenticeship Reinforcement Learning for Control
  46. Why Reinforcement Learning is Important
  47. Peekaboom
  48. Not goal metrics
  49. Interesting papers at ACL
  50. Six Months
  51. What Learning Theory might do
  52. Text Entailment at AAAI
  53. “Sister Conference” presentations
  54. AAAI blog
  55. Thinking the Unthought
  56. The Limits of Learning Theory
  57. The Health of COLT
  58. The Role of Impromptu Talks
  59. Not EM for clustering at COLT
  60. A COLT paper
  61. The cross validation problem: cash reward
  62. Languages of Learning
  63. Lower Bounds for Learning Reductions
  64. Reopening RL->Classification
  65. Wikis for Summer Schools and Workshops
  66. Workshops are not Conferences
  67. Question: “When is the right time to insert the loss function?”
  68. Exact Online Learning for Classification
  69. Maximum Margin Mismatch?
  70. Bad ideas
  71. Running A Machine Learning Summer School
  72. What is the right form of modularity in structured prediction?
  73. A Short Guide to PhD Graduate Study
  74. Regret minimizing vs error limiting reductions
  75. NIPS
  76. Math on the Web
  77. Visa Casualties
  78. Learning Reductions are Reductionist
  79. Don’t mix the solution into the problem
  80. Conference attendance is mandatory
  81. Reviewing techniques for conferences
  82. Science Fiction and Research
  83. DARPA project: LAGR
  84. To calibrate or not?
  85. Embeddings: what are they good for?
  86. Advantages and Disadvantages of Bayesian Learning
  87. New Blog: [Lowerbounds,Upperbounds]
  88. Dynamic Programming Generalizations and Their Use
  89. Which Assumptions are Reasonable?
  90. Families of Learning Theory Statements
  91. Is the Goal Understanding or Prediction?
  92. Fast SVMs
  93. Structured Regret Minimization
  94. Grounds for Rejection
  95. Basic computer science research takes a hit
  96. The Producer-Consumer Model of Research
  97. What can Type Theory teach us about Machine Learning?
  98. Academic Mechanism Design
  99. Open Problems for Colt
  100. The Role of Workshops
  101. Active learning
  102. Research Styles in Machine Learning
  103. Binomial Weighting
  104. Going all the Way, Sometimes
  105. The State of Tight Bounds
  106. Avoiding Bad Reviewing
  107. Breaking Abstractions
  108. Bad Reviewing
  109. Fast Physics for Learning
  110. Funding Research
  111. The Big O and Constants in Learning
  112. Prior, “Prior” and Bias
  113. Regularization
  114. Antilearning: When proximity goes bad
  115. Problem: Reductions and Relative Ranking Metrics
  116. Why Papers?
  117. Problem: Online Learning
  118. Solution: Reinforcement Learning with Classification
  119. Problem: Reinforcement Learning with Classification
  120. Problem: Cross Validation
  121. At One Month
  122. Machine learning reading groups
  123. Loss Functions for Discriminative Training of Energy-Based Models
  124. What it means to do research.
  125. Learning Research Programs
  126. ESPgame and image labeling
  127. Clever Methods of Overfitting
  128. ROC vs. Accuracy vs. AROC
  129. Conferences, Dates, Locations
  130. Intuitions from applied learning
  131. Some Links
  132. The State of the Reduction
  133. JMLG
  134. Learning Theory, by assumption
  135. Paper Deadlines
  136. Kolmogorov Complexity and Googling
  137. NIPS: Online Bayes
  138. Watchword: Loss
  139. Watchword: Assumption
  140. Learning Complete Problems
  141. Watchword: Probability
  142. Summer Schools
  143. Holy grails of machine learning?
  144. The Humanloop Spectrum of Machine Learning
  145. Why I decided to run a weblog.