A new estimator improves training of probabilistic models with latent Gaussian variables.
problem Improving gradient estimation for models with latent Gaussian variables.
method Rao-Blackwellised Reparameterisation Gradients (R2-G2)
result R2-G2 consistently yields better performance in models with multiple applications of the reparameterisation trick.
Probabilistic programming has emerged as a powerful paradigm in statistics, applied science, and machine learning: by decoupling modelling from inference, it promises to allow modellers to directly reason about the processes generating data. However, the performance of inference algorithms can be dramatically affected …
This Ph.D. thesis is devoted to the constructions of Lagrangian formulation on Finsler and Kawaguchi manifolds. While Finsler geometry is a natural extension of Riemannian geometry, Kawaguchi geometry is the extension of Finsler geometry to higher order derivatives and to k-dimensional parameter space. The latter exten…
Improved particle filters for estimating model parameters using differentiable resampling.
problem Inability to differentiate sampling and resampling steps in particle filters.
method Extended reparameterisation trick to include stochastic input, enabling differentiation. Used p-MCMC and NUTS for parameter estimation.
result NUTS improves mixing of Markov chain and produces more accurate results in less time.
Powerpropagation makes neural networks inherently sparse.
problem Training sparse neural networks to reduce computational footprint and model size.
method Introduces a new weight-parameterisation technique exploiting gradient descent dynamics.
result Models trained with Powerpropagation have a higher density of zero weights, allowing for more efficient pruning.
We address the problem of simultaneously learning a k-means clustering and deep feature representation from unlabelled data, which is of interest due to the potential of deep k-means to outperform traditional two-step feature extraction and shallow-clustering strategies. We achieve this by developing a gradient-estimat…
We introduce a family of pairwise stochastic gradient estimators for gradients of expectations, which are related to the log-derivative trick, but involve pairwise interactions between samples. The simplest example of our new estimator, dubbed the fundamental trick estimator, is shown to arise from either a) introducin…
Wide Bayesian neural networks have a simpler weight posterior, leading to faster MCMC sampling.
problem Sampling from the posterior of wide Bayesian neural networks is challenging.
method Introducing repriorisation, a data-dependent reparameterisation that simplifies the posterior distribution.
result The repriorisation map accelerates MCMC sampling, achieving up to 50x higher effective sample size.
Hierarchical IBP model for Bayesian neural networks in continual learning.
problem Resource allocation in continual learning with dynamic network complexity.
method Indian Buffet process (IBP) and Hierarchical-IBP (H-IBP) priors for structure learning, online variational inference with reparameterization.
result Our model effectively learns the number of weights in each layer, overcoming overfitting and underfitting.
Deep learning for missing data with flexible modelling.
problem Missing data with missingness dependent on the missing values.
method Deep latent variable models with a neural network for missingness pattern modeling, importance-weighted variational inference, and stochastic gradients.
result Explicitly modelling missingness improves inference on various data sets.
The Gaussian process state space model (GPSSM) is a non-linear dynamical system, where unknown transition and/or measurement mappings are described by GPs. Most research in GPSSMs has focussed on the state estimation problem, i.e., computing a posterior of the latent state given the model. However, the key challenge in…
Integrable nets described with curvature relations to pseudospherical surfaces.
problem Describing integrable curve nets and their geometric properties.
method Overview of second-order invariants, specific example of concordant nets, and construction of pseudospherical surfaces.
result Concordant Chebyshev nets correspond to pairs of pseudospherical surfaces.
SNEPPPs use squared neural networks to efficiently model Poisson point processes.
problem Efficiently modeling Poisson point processes with flexibility.
method Parameterizing intensity function with squared norm of a two-layer neural network.
result Closed-form integration of intensity function for quadratic time computation.
This paper improves MADDPG's performance in discrete grid-world scenarios.
problem MADDPG struggles with discrete action spaces in grid-world tasks.
method Integrates alternative Gumbel-Softmax estimators into MADDPG.
result One alternative estimator significantly improves performance in grid-world tasks.
A new neural topic model using optimal transport improves document representation and topic coherence.
problem Challenges in achieving good document representation and coherent/diverse topics in existing NTMs.
method Proposes a neural topic model via optimal transport, learning topic distribution by minimising OT distance to document word distributions.
result Significantly outperforms state-of-the-art NTMs on discovering coherent and diverse topics.