Accepted papers, 2011
The Big Learning 2011 programme accepted 32 papers, arranged across the two workshop days. Each day had contributed talks and spotlights. The titles below connect hardware and software questions with statistical inference, evaluation and optimisation, showing the range of subjects that appeared together at the Sierra Nevada meeting.
The four groups follow the programme’s Day 1 and Day 2 divisions. The edition introduction explains the two-day setting, and the invited-talk page supplies the complementary speaker programme. The paper authors remain in the order attached to each accepted title.
Arrays, languages and parallel resources
The 2011 description at NeurIPS placed algorithms, systems and tools within one workshop. The accepted titles reflect that relationship. The first day included a common GPU array, GPU-based probabilistic inference and Theano. For background on the latter software, the Theano Development Team’s 2016 article described a Python library for defining, optimising and evaluating expressions involving multidimensional arrays. That software description supplies context for the array and computation vocabulary in the programme.
Several titles concerned languages, libraries and frameworks. They appeared alongside titles about models and learning procedures, bringing questions of expression and implementation into the same list. A language or library title points towards the organisation of computational work, while an inference title points towards the statistical task. The hardware topic page develops the distinction between processor choices and the software that presents their operations.
Evaluation and uncertainty
Cross-validation and the bootstrap appeared among the contributed titles. Wikipedia’s cross-validation explanation describes the assessment of generalisation using data beyond that used for fitting. Its bootstrap explanation describes resampling to estimate an estimator’s distribution. These are distinct evaluation ideas. Their presence in the programme makes assessment part of the title-level map of big learning, alongside the work of constructing and updating a model.
Other titles named Bayesian inference, Hamiltonian Monte Carlo, graphical models and stochastic optimisation. The two-day arrangement therefore placed several forms of statistical computation beside parallel implementation subjects. The inference topic page introduces these families of terms. Their appearance here identifies subjects represented at the meeting; a method’s assumptions and detailed evidence remain questions for its own research account.
Representations and learning objectives
Hashing titles sit beside graph classification, kernel clustering, semantic parsing, ranking and matrix factorisation. These titles point to different representations and learning tasks. Some name a data structure or a way of working with many features; others name the objective of classification, clustering or ranking. The programme also included distributed and remote-data subjects, connecting those mathematical choices with the location of computation.
The printed groups distinguish longer contributed presentations from spotlights on each day. Those divisions describe the programme format rather than a ranking of research. Readers can follow a subject across groups: inference terms appear in several places, as do software and optimisation terms. The hashing page and optimisation page provide further vocabulary for reading across the complete list.
Day 1 - December 16th, 2011
Day 1 Contributed Talks
- A Common GPU n-Dimensional Array for Python and C
- Parallelizing the Training of the Kinect Body Parts Labeling Algorithm
- Fast Cross-Validation via Sequential Analysis
- Bootstrapping Big Data
Day 1 Spotlights
- Theano: Deep Learning on GPUs with Python
- Torch7: A Matlab-like Environment for Machine Learning
- Parallel Algorithms for GPU accelerated Probabilistic Inference
- Multiple Hash Functions for Learning
- b-Bit Minwise Hashing for Large-Scale Learning
- Combining approximate inference methods for efficient learning on large computer clusters
- Very Large Scale Bayesian Inference Using MCDB
- Learning Updatable Classifiers from Remote Data
- Large-scale Bayesian ICA with application to separation of the cosmic microwave background
- A Randomized Gossip-based Algorithm for Classification on Peer-to-Peer Networks
- Distributed Unsupervised Semantic Parsing of Large-Scale Document Corpora
- Graph Based Classification of Content and Users in BitTorrent
Day 2 - December 17th, 2011
Day 2 Contributed Talks
- The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo
- Randomized Smoothing for (Parallel) Stochastic Optimization
- Block Splitting for Large-Scale Distributed Learning
- Large-Scale Matrix Factorization with Distributed Stochastic Gradient Descent
Day 2 Spotlights
- Towards Asynchronous Distributed MCMC Inference for Large Graphical Models
- Solving Large Scale Linear SVM with Distributed Block Minimization
- Scalable Kernel k-Means via Centroid Approximation
- Semi-Supervised Learning with Cover Trees
- A Theoretical Analysis of a Warm Start Technique
- Scalable Stochastic Gradient Descent with Improved Confidence
- Distributed Learning-to-Rank on Streaming Data using Alternating Direction Method of Multipliers
- Machine learning in ScalOps, a higher order cloud computing language
- MLPACK: A Scalable C++ Machine Learning Library
- GURLS: a Toolbox for Large Scale Multiclass Learning
- Tools and Frameworks for Big Learning in Scala: Leveraging the Language for High Productivity and Performance
- A Reliable Effective Terascale Linear Learning System