Time and place: 2–3:30pm on roughly alternate Mondays, IF 1.16.
Date | Person | Paper(s) |
---|---|---|
20 Jan | James | Fast and Accurate Least-Mean-Squares Solvers |
3 Feb | Iain | HNSW for approximate nearest neighbours |
17 Feb | Tiffany | Partitioned integrators for thermodynamic parameterization of neural networks |
9 Mar | Conor | MC Gradient Estimation in ML (Sections 4, 5, 7), Concrete Distribution & Gumbel-Softmax (reparameterization trick for discrete variables), REBAR (control variates for score function estimator using concrete/GS) |
31 Mar | Artur | Contrastive methods for representation learning. A Simple Framework for Contrastive Learning of Visual Representations, Representation Learning with Contrastive Predictive Coding, Data-Efficient Image Recognition with Contrastive Predictive Coding |
15 Apr | Asa | Calibration under dataset shift Can you trust your model's uncertainty? Evaluating predictive uncertainty under dataset shift |
29 Apr | James | SUMO: Unbiased Estimation of Log Marginal Probability for Latent Variable Models |
4 May | ||
18 May | ||
1 June | ||
15 June | ||
29 June |
Reset the list for 2020. Brainstorm of ideas, not a schedule:
- ...
- ...
-
HNSW for approximate nearest neighbours does well in some benchmarks. (Although FAISS offers some discussion of the choices for different regimes.)
-
nmslib implements HNSW, and has a reading list.
-
Any NN-graph methods? Classics like isomap and LLE. Discuss how nearest neighbours can fit into more recent methods?
Anything that helps making practical choices, or gives insight into what choices matter?
Batch size:
-
Which Algorithmic Choices Matter at Which Batch Sizes? Insights From a Noisy Quadratic Model
-
Measuring the Effects of Data Parallelism on Neural Network Training
Learning rates:
- ...
(Not continuous-like data, such as quantized images.)
Fast generation of high-dim patterns?
- Non-Autoregressive Neural Machine Translation
- Latent-Variable Non-Autoregressive Neural Machine Translation with Deterministic Inference Using a Delta Posterior
- RBMs?
Methods for latent variable models. Concrete distribution, REBAR/RELAX...?
-
More on exact estimation with MCMC?
-
Unbiased scalable softmax optimization (although critical review
-
Efficient Optimization of Loops and Limits with Randomized Telescoping Sums
-
...?
- ...?
- Graph Neural Networks?