Jake Abernethy

Conference Papers

• A Stochastic View of Optimal Regret through Minimax Duality
  Jacob Abernethy, Alekh Agarwal, Peter Bartlett and Alexander Rakhlin
  [PDF - Draft] - Conference on Learning Theory (June 2009)
• Beating the Adaptive Bandit with High Probability
  Jacob Abernethy and Alexander Rakhlin
  [PDF - Draft] - Conference on Learning Theory (June 2009)
• Optimal Strategies and Minimax Lower Bounds for Online Convex Games
  Jacob Abernethy, Peter Bartlett, Alexander Rakhlin and Ambuj Tewari
  [PDF] - Conference on Learning Theory (July 2008)
• Competing in the Dark: An Efficient Algorithm for Bandit Linear Optimization
  Jacob Abernethy, Elad Hazan and Alexander Rakhlin
  Best Student Paper Award
  [PDF] - Conference on Learning Theory (July 2008)
• Optimal Strategies from Random Walks
  Jacob Abernethy, Manfred Warmuth and Joel Yellin
  [PDF] - Conference on Learning Theory (July 2008)
• Webspam Identification Through Content and Hyperlinks
  Jacob Abernethy, Olivier Chapelle and Carlos Castillo
  [PDF] - AirWeb 2008, Adversarial Information Retrieval on the Web (April 2008)
• Online Discovery of Similarity Mappings
  Alexander Rakhlin, Jacob Abernethy, Peter Bartlett
  [PDF] - International Conference on Machine Learning (June 2007)
• Multitask Learning with Expert Advice
  Jacob Abernethy, Peter Bartlett, Alexander Rakhlin
  [PDF] - Conference on Learning Theory (June 2007)
• Continuous Experts and the Binning Algorithm
  Jacob Abernethy, John Langford, Manfred Warmuth
  [PDF] - Conference on Learning Theory (June 2006)

Journal Papers

• A New Approach to Collaborative Filtering: Operator Estimation with Spectral Regularization
  Jacob Abernethy, Francis Bach, Theodoros Evgeniou, Jean-Philippe Vert
  [PDF] - Journal of Machine Learning Research, Vol. 10(Mar):803--826, 2009
• Eliciting Consumer Preferences Using Robust Adaptive Choice Questionnaires
  Jacob Abernethy, Theodoros Evgeniou, Olivier Toubia, Jean-Philippe Vert
  [PDF] - IEEE Transactions on Knowledge and Data Engineering, Vol. 20, No. 2
• Graph Regularization Methods for Web Spam Detection
  Jacob Abernethy, Olivier Chapelle and Carlos Castillo
  [Soon] - Machine Learning Journal (Accepted for Publication, November 2008)

Open Problems

• Minimax Games with Bandits
  Jacob Abernethy and Manfred Warmuth
  [PDF] - Conference on Learning Theory (June 2009)
• An Efficient Bandit Algorithm for T^(1/2)-Regret in Online Multiclass Prediction?
  Jacob Abernethy and Alexander Rakhlin
  [PDF] - Conference on Learning Theory (June 2009)

Technical Reports and Notes

• Low-rank matrix factorization with attributes
  Jacob Abernethy, Francis Bach, Theodoros Evgeniou and Jean-Philippe Vert
  [PDF] - arXiv:cs/0611124 (November 2006)

• Optimal Randomized Paging with Follow the Perturbed Leader with Henry Lin
  Paging is one of the simplest online decision problems: given a cache of size k and an unknown sequence of page requests, every time a cache miss occurs, which element of the cache should I evict? The well-known "Marking Algorithm" can be shown to obtain a 2 log k competitive ratio, while an algorithm by Sleator and McGeoch from 1991 gets the optimal log k rate, although the algorithm is complex and the analysis quite involved. We show that a simple technique of "guess the future" obtains the same optimal competitive ratio but with a somewhat more intuitive algorithm and analysis. Our approach may be applicable to more difficult online problems such as k-server and metrical task systems.
• Repeated Game Playing Against a Budgeted Adversary with Manfred Warmuth
  Assume you will play a zero-sum game against an adversary, but where the adversary has some budget on how many times he can play each strategy. The budget can be any monotonic function of the action counts, and once the budget has been reached the game is over. We show that, for a large class of payoff matrices M, the value of this game, as well as the optimal strategies of both players, can be described by a simple random walk with transitions probabilities depending on the matrix M. Indeed, we show that such repeated games are essentially isomorphic to the "prediction with expert advice" problem.

• I was recently awarded the 2008-2009 Yahoo! PhD Fellowship. Thank you Yahoo!!
• I'm the current president of the CSGSA, the organization for graduate students in Berkeley's Computer Science department. We run weekly social events, take students out for beer, cajole prospective students into attending, vet potential faculty members, play softball, prepare younger students for tests, and several other useful activities. The CSGSA is made up of an awesome team of officers, who do the real work anyway.
• A talk at Google on the subject of Web spam classification. This follows some work done with Olivier Chapelle and Carlos Castillo while I was a summer intern at Yahoo! Research.
• A photo taken during a juggling show at MIT. Don't ask why they called me Sinbad.
• A reporter caught me juggling on a ladder.
• In a previous life, I used to do comedy juggling shows. I managed to win Funniest Student at UMass while studying there, and later got the chance to open for the comedian Sinbad and later Dave Chappelle. The Daily Hampshire Gazette ran a flattering article about my show right before the Sinbad appearance.
• I love playing Racquetball.
• I rode a bicycle from Boston to San Francisco during the summer of 2002 with two friends, Keith Vanderlinde and Damien Burke. Keith had the good sense to document the trip.
• My mom, an expert on Anthrax and Anthrax Vaccine, has a very popular informational Anthrax Vaccine Web site and blog.
• I used to help manage MEET: Middle East Education through Technology, an educational initiative aimed at bringing together Israeli and Palestinian youth. This project has brought me to Jerusalem three times.