arXiv subjects

James Wallbridge

Publications and source records attributed to James Wallbridge.

7 recordsLinked to original sources

arXiv2021

Geometry of Program Synthesis

We re-evaluate universal computation based on the synthesis of Turing machines. This leads to a view of programs as singularities of analytic varieties or, equivalently, as phases of the Bayesian posterior of a synthesis problem. This new point of view reveals unexplored directions of research in program synthesis, of which neural networks are a subset, for example in relation to phase transitions, complexity and generalisation. We also lay the empirical foundations for these new directions by reporting on our implementation in code of some simple experiments.

cs.LG↗

arXiv2020

Transformers for Limit Order Books

We introduce a new deep learning architecture for predicting price movements from limit order books. This architecture uses a causal convolutional network for feature extraction in combination with masked self-attention to update features based on relevant contextual information. This architecture is shown to significantly outperform existing architectures such as those using convolutional networks (CNN) and Long-Short Term Memory (LSTM) establishing a new state-of-the-art benchmark for the FI-2010 dataset.

q-fin.CP↗

arXiv2019

Jets and differential linear logic

We prove that the category of vector bundles over a fixed smooth manifold and its corresponding category of convenient modules are models for intuitionistic differential linear logic. The exponential modality is modelled by composing the jet comonad, whose Kleisli category has linear differential operators as morphisms, with the more familiar distributional comonad, whose Kleisli category has smooth maps as morphisms. Combining the two comonads gives a new interpretation of the semantics of differential linear logic where the Kleisli morphisms are smooth local functionals, or equivalently, smooth partial differential operators, and the codereliction map induces the functional derivative. This points towards a logic and hence computational theory of non-linear partial differential equations and their solutions based on variational calculus.

cs.LO↗

arXiv2019

Logic and the $2$-Simplicial Transformer

We introduce the $2$-simplicial Transformer, an extension of the Transformer which includes a form of higher-dimensional attention generalising the dot-product attention, and uses this attention to update entity representations with tensor products of value vectors. We show that this architecture is a useful inductive bias for logical reasoning in the context of deep reinforcement learning.

cs.LG↗

arXiv2018

Homotopy theory in a quasi-abelian category

We prove that the category of dg-modules and dg-algebras in a Grothendieck quasi-abelian category are endowed with a Quillen model structure. This allows some flexibility in setting up a theory of derived algebraic geometry in the infinite dimensional setting. For example, the category of complete bornological vector spaces, or equivalently, convenient vector spaces, is a Grothendieck quasi-abelian category. Closely related is the Grothendieck quasi-abelian category of ind-Banach spaces whose associated model category is shown to be Quillen equivalent. Applications include the Chevalley-Eilenberg resolution and the Koszul resolution of a commutative monoid object in a Grothendieck quasi-abelian category. These can be used for the calculation of derived quotients by an infinite dimensional Lie algebra and derived intersections respectively.

math.AT↗

arXiv2017

Tannaka duality over ring spectra

We prove a Tannaka duality theorem for $(\infty,1)$-categories. This is a duality between certain derived group stacks, or more generally certain derived gerbes, and symmetric monoidal $(\infty,1)$-categories endowed with particular structure. This duality theorem is defined over commutative ring spectra and subsumes the classical statement. We show how the classical theory, and its extension over arbitrary rings, arises as a special case of our more general theory. The application to perfect complexes is explored.

math.AG↗

arXiv2017

Derived smooth stacks and prequantum categories

The Weil-Kostant integrality theorem states that given a smooth manifold endowed with an integral complex closed 2-form, then there exists a line bundle with connection on this manifold with curvature the given 2-form. It also characterises the moduli space of line bundles with connection that arise in this way. This theorem was extended to the case of p-forms by Gajer in [Ga]. In this paper we provide a generalization of this theorem where we replace the original manifold by a derived smooth Artin stack. Our derived Artin stacks are geometric stacks on the étale (\infty,1)-site of affine derived smooth manifolds. We introduce the notion of a n-shifted p-preplectic derived smooth Artin stack in analogy with the algebraic case constructed by Pantev-Toën-Vaquié-Vezzosi in [PTVV]. This is a derived smooth Artin stack endowed with a complex closed (p+1)-form which has been cohomologically shifted by degree n. It is a far reaching generalization of a p-preplectic manifold which includes orbifolds and other highly singular objects. We then show that when its n-shifted p-preplectic form is integral, then there exists a (p+n-1)-gerbe with p-connection data and curvature corresponding to the original p-preplectic form. We also provide the characterization of the moduli stack of gerbes with connections arising in this context. We construct a canonical functor from the (\infty,1)-category of integral n-shifted p-preplectic derived smooth Artin stacks to the (\infty,1)-category of linear (\infty,p+n-1)-categories. When n=0 and p=1, this functor can be thought of like a cohomology functor in that it associates to a derived presymplectic smooth Artin stack a linear invariant in the form of a differential graded module. In the general case we obtain higher prequantum categories which requires the machinery of linear (\infty,n)-categories.

math.SG↗