The 2013 workshop day
The 2013 Big Learning workshop met on Monday 9 December at Lake Tahoe, in Harveys Emerald Bay B. Its day combined invited talks, three contributed presentations, poster discussions and a tutorial on Vowpal Wabbit. Morning and afternoon sessions surrounded a long midday interval, so the programme alternated talks addressed to the whole room with time for discussion around individual papers.
The title, Big Learning: Advances in Algorithms and Data Management, identified the meeting's central relationship. The NeurIPS description of the 2013 workshop set out a conversation between machine-learning and database research. The timetable put that conversation beside work on distributed computation, clustering and probabilistic inference. The closing invited talk was cancelled; it did not form part of the completed programme.
Morning: systems and contributed work
Poster setup opened the day before the first pair of invited talks. Their titles placed data work and computation across machines at the start of the programme. A combined poster session and coffee break followed, separating those longer presentations from the morning's three contributed talks. That arrangement gave the day two different scales of discussion: a shared sequence of talks and a collection of smaller conversations around papers.
The contributed titles named a parameter server, distributed principal component analysis and clustering, and parallel inference for a Dirichlet process. These are different kinds of problem. A system for sharing model parameters concerns the movement and coordination of model state; a dimensionality-reduction or clustering method concerns how a data analysis is divided; an inference method concerns the relationship between a statistical procedure and its parallel execution. The titles make those subjects visible without establishing the papers' methods or conclusions.
| Time | Activity |
|---|---|
| 7:00–7:30 AM | Poster Setup |
| 7:30–8:15 AM | The Thorn in the Side of Big Data: Too Few ArtistsChris Re |
| 8:15–9:00 AM | Scale-out Beyond Map-ReduceRaghu |
| 9:00–9:30 AM | Poster Session + Coffee Break |
| 9:30–9:50 AM | Parameter Server for Distributed Machine LearningAlex Smola |
| 9:50–10:10 AM | Distributed PCA and k-Means ClusteringYingyu Liang |
| 10:10–10:30 AM | Pitfalls in the use of Parallel Inference for the Dirichlet ProcessYarin Gal |
| Midday interval | Poster Session + Lunch + Skiing |
| 1:30–3:00 PM | Tutorial on Vowpal WabbitJohn Langford |
A tutorial during the midday interval
The morning session ended with a poster session, lunch and skiing on the programme. During that break, John Langford presented the Vowpal Wabbit tutorial from 1:30 PM to 3:00 PM. The Machine Learning (Theory) account of the tutorial gave that time and the Emerald Bay B room. The tutorial was a distinct session within the day, rather than an additional contributed paper or a cancelled presentation.
A tutorial has a different purpose from a paper talk. It can introduce a tool's concepts and working conventions across a longer block of time. That role fitted a workshop concerned with the connection between learning algorithms and their implementation. The 2013 invited-talks page places the tutorial beside the invited programme, while the accepted-papers page supplies the complete contributed and poster lists.
Afternoon: dataflow and statistical computation
The afternoon began with two invited talks. Their titles connected dataflow with the computational and statistical interface. A poster session and coffee break followed. The final scheduled invited slot carried a cancellation, which matters when distinguishing the planned sequence from the talks that took place. The day therefore contained four completed invited talks, the three morning contributed talks and the midday tutorial.
The poster periods were an integral part of this arrangement. They offered space beside the fixed speaking slots, allowing several papers to be discussed during the same interval. The sequence of talks, breaks and posters also shows why a workshop programme cannot be understood solely as a list of invited speakers. Contributed work and informal discussion occupied substantial parts of the day, alongside the longer presentations that framed its subjects.
| Time | Activity |
|---|---|
| 3:30–4:15 PM | Timely dataflow in NaiadDerek Murray |
| 4:15–5:00 PM | On the Computational and Statistical Interface and Big DataMichael Jordan |
| 5:00–5:45 PM | Poster Session + Coffee Break |
| 5:45–6:30 PM | Big Empathy: Growing Up (Canceled)Joe HellersteinCancelled |
The earlier editions used related formats
The 2011 edition occupied 16–17 December in Sierra Nevada, Spain. Both days had morning and afternoon sessions. Invited talks appeared alongside contributed talks, poster spotlights and poster sessions; the tutorials covered Vowpal Wabbit on the first day and GraphLab 2.0 on the second. A spotlight and a poster served different functions: a short presentation drew attention to a paper, while the poster period made room for direct discussion.
The 2012 edition met on 8 December at Lake Tahoe. Its morning and afternoon sessions included three invited speakers and four contributed talks, separated by poster periods and a midday break. The NeurIPS description of the 2012 workshop organised its interests around data, models and algorithms, applications, and tools and systems. Those categories describe the breadth of the day rather than four separate timetabled tracks.
Across the editions, the programme joined algorithmic questions to computing choices. The formats varied in duration and emphasis, but talks, posters and tutorials each made a different kind of exchange possible. The key-dates page follows the calendars that led to these meetings, and the glossary explains the presentation formats alongside the technical terms that recur in the programmes.