L4

A defining feature of robotics today is the use of learning and autonomy in the inner loop of systems that are actually being deployed in the real world, e.g., in autonomous driving or medical robotics. While it is clear that useful autonomous systems must learn to cope with a dynamic environment, requiring architectures that address the richness of the worlds in which such robots must operate, it is also equally clear that ensuring the safety of such systems is the single biggest obstacle preventing scaling up of these solutions. I will discuss an approach to system design that aims at addressing this problem by incorporating programmatic structure in the network architectures being used for policy learning. I will discuss results from two projects in this direction.

Firstly, I will present the perceptor gradients algorithm – a novel approach to learning symbolic representations based on the idea of decomposing an agent’s policy into i) a perceptor network extracting symbols from raw observation data and ii) a task encoding program which maps the input symbols to output actions. We show that the proposed algorithm is able to learn representations that can be directly fed into a Linear-Quadratic Regulator (LQR) or a general purpose A* planner. Our experimental results confirm that the perceptor gradients algorithm is able to efficiently learn transferable symbolic representations as well as generate new observations according to a semantically meaningful specification.

Next, I will describe work on learning from demonstration where the task representation is that of hybrid control systems, with emphasis on extracting models that are explicitly verifi able and easily interpreted by robot operators. Through an architecture that goes from the sensorimotor level involving fitting a sequence of controllers using sequential importance sampling under a generative switching proportional controller task model, to higher level modules that are able to induce a program for a visuomotor reaching task involving loops and conditionals from a single demonstration, we show how a robot can learn tasks such as tower building in a manner that is interpretable and eventually verifiable.

References:

1. S.V. Penkov, S. Ramamoorthy, Learning programmatically structured representations with preceptor gradients, In Proc. International Conference on Learning Representations (ICLR), 2019. http://rad.inf.ed.ac.uk/data/publications/2019/penkov2019learning.pdf

2. M. Burke, S.V. Penkov, S. Ramamoorthy, From explanation to synthesis: Compositional program induction for learning from demonstration, https://arxiv.org/abs/1902.10657
 

Deep generative models provide powerful tools for fitting difficult distributions such as modelling natural images. But many of these methods, including  variational autoencoders (VAEs) and generative adversarial networks (GANs), can be notoriously difficult to fit.

One well-known problem is mode collapse, which means that models can learn to characterize only a few modes of the true distribution. To address this, we introduce VEEGAN, which features a reconstructor network, reversing the action of the generator by mapping from data to noise. Our training objective retains the original asymptotic consistency guarantee of GANs, and can be interpreted as a novel autoencoder loss over the noise.

Second, maximum mean discrepancy networks (MMD-nets) avoid some of the pathologies of GANs, but have not been able to match their performance. We present a new method of training MMD-nets, based on mapping the data into a lower dimensional space, in which MMD training can be more effective. We call these networks Ratio-based MMD Nets, and show that somewhat mysteriously, they have dramatically better performance than MMD nets.

A final problem is deciding how many latent components are necessary for a deep generative model to fit a certain data set. We present a nonparametric Bayesian approach to this problem, based on defining a (potentially) infinitely wide deep generative model. Fitting this model is possible by combining variational inference with a Monte Carlo method from statistical physics called Russian roulette sampling. Perhaps surprisingly, we find that this modification helps with the mode collapse problem as well.

Generative adversarial networks (GANs) use neural networks as generative models, creating realistic samples that mimic real-life reference samples (for instance, images of faces, bedrooms, and more). These networks require an adaptive critic function while training, to teach the networks how to move improve their samples to better match the reference data. I will describe a kernel divergence measure, the maximum mean discrepancy, which represents one such critic function. With gradient regularisation, the MMD is used to obtain current state-of-the art performance on challenging image generation tasks, including 160 × 160 CelebA and 64 × 64 ImageNet. In addition to adversarial network training, I'll discuss issues of gradient bias for GANs based on integral probability metrics, and mechanisms for benchmarking GAN performance.

Data science has become a topic of great interest lately and has triggered new widescale research activities around efficientl first order methods for optimisation and Bayesian sampling. The National Physical Laboratory is addressing some of these challenges with particular focus on  robustness and confidence in the solution.  In this talk, I will present some problems and recent results concerning i. robust learning in the presence of outliers based on the Median of Means (MoM) principle and ii. stability of the solution in super-resolution (joint work with A. Thompson and B. Toader).

Mysteries of isolated horizons: the Near Horizon Geometry equation, geometric characterizations of the non-extremal Kerr horizon, spacetimes foliated by non-expanding horizons.

3-dimensional null surfaces  that are  Killing horizons to the second order  are  considered. They are embedded in 4-dimensional spacetimes that satisfy the vacuum Einstein equations with arbitrary cosmological constant. Internal geometry of 2-dimensional cross sections of  the horizons  consists of induced metric tensor and a rotation 1-form potential. It is subject to the type D equation. The equation is interesting from the both, mathematical and physical points of view. Mathematically it  involves  geometry, holomorphic structures and algebraic topology.  Physically, the equation knows the secrete of black holes: the only  axisymmetric solutions on topological sphere  correspond  to the the Kerr / Kerr-de Sitter / Kerr-anti-de-Sitter non-extremal black holes or to the near horizon limit  of the extremal ones.  In the case of bifurcated  horizons the type D equation implies another spacial  symmetry. In this way the axial symmetry may be ensured without the rigidity theorem. The type D equation does not allow rotating horizons of topology different then that of the  sphere (or its quotient). That completes a new local non-her theorem. The type D equation is also  an integrability condition for the  Near Horizon Geometry equation and leads to new results on the solution existence issue.
 

I will talk about the implications of UV completion of quantum gravity on the low energy spectrums. I will introduce the constraints on low-energy effective theory due to unitarity and analyticity of scattering amplitudes, in particular an infinite set of new unitarity constraints on the forward-limit limit of four-point scattering amplitudes due to the work of Arkani-Hamed et al. In three dimensions, we find the constraints imply that light states with charge-to-mass ratio z greater than 1 can only be consistent if there exists other light states, preferably neutral. Applied to the 3D Standard Model like spectrum, where the low energy couplings are dominated by the electron with z \sim 10^22, this provides a novel understanding of the need for light neutrinos.

Many toric degenerations and integrable systems of the Grassmannians Gr(2, n) are described by trees, or equivalently subdivisions of polygons. These degenerations can also be seen to arise from the cones of the tropicalisation of the Grassmannian. In this talk, I focus on particular combinatorial types of cones in tropical Grassmannians Gr(k,n) and prove a necessary condition for such an initial degeneration to be toric. I will present several combinatorial conjectures and computational challenges around this problem.  This is based on joint works with Kristin Shaw and with Oliver Clarke.

Enumerative algebraic geometry counts the solutions to certain geometric constraints. Numerical algebraic geometry determines these solutions for any given 
instance. This lecture illustrates how these two fields complement each other, especially in the light of emerging new applications. We start with a gem from
the 19th century, namely the 3264 conics that are tangent to five given conics in the plane. Thereafter we turn to current problems in statistics, with focus on 
maximum likelihood estimation for linear Gaussian covariance models.
 

I will review the fishnet model, which is an integrable scalar QFT, obtained by an extreme gamma deformation of N=4 super Yang-Mills. The theory has a peculiar perturbative expansion in which many quantities at a fixed loop order are given by a single Feynman diagram. This feature allows the reinterpretation of Feynman loop integrals as integrable systems.