-
Active Inference as a Model of Agency
Authors:
Lancelot Da Costa,
Samuel Tenka,
Dominic Zhao,
Noor Sajid
Abstract:
Is there a canonical way to think of agency beyond reward maximisation? In this paper, we show that any type of behaviour complying with physically sound assumptions about how macroscopic biological agents interact with the world canonically integrates exploration and exploitation in the sense of minimising risk and ambiguity about states of the world. This description, known as active inference,…
▽ More
Is there a canonical way to think of agency beyond reward maximisation? In this paper, we show that any type of behaviour complying with physically sound assumptions about how macroscopic biological agents interact with the world canonically integrates exploration and exploitation in the sense of minimising risk and ambiguity about states of the world. This description, known as active inference, refines the free energy principle, a popular descriptive framework for action and perception originating in neuroscience. Active inference provides a normative Bayesian framework to simulate and model agency that is widely used in behavioural neuroscience, reinforcement learning (RL) and robotics. The usefulness of active inference for RL is three-fold. \emph{a}) Active inference provides a principled solution to the exploration-exploitation dilemma that usefully simulates biological agency. \emph{b}) It provides an explainable recipe to simulate behaviour, whence behaviour follows as an explainable mixture of exploration and exploitation under a generative world model, and all differences in behaviour are explicit in differences in world model. \emph{c}) This framework is universal in the sense that it is theoretically possible to rewrite any RL algorithm conforming to the descriptive assumptions of active inference as an active inference algorithm. Thus, active inference can be used as a tool to uncover and compare the commitments and assumptions of more specific models of agency.
△ Less
Submitted 23 January, 2024;
originally announced January 2024.
-
Additive Invariants of Open Petri Nets
Authors:
Benjamin Merlin Bumpus,
Sophie Libkind,
Jordy Lopez Garcia,
Layla Sorkatti,
Samuel Tenka
Abstract:
We classify all additive invariants of open Petri nets: these are $\mathbb{N}$-valued invariants which are additive with respect to sequential and parallel composition of open Petri nets. In particular, we prove two classification theorems: one for open Petri nets and one for monically open Petri nets (i.e. open Petri nets whose interfaces are specified by monic maps). Our results can be summarize…
▽ More
We classify all additive invariants of open Petri nets: these are $\mathbb{N}$-valued invariants which are additive with respect to sequential and parallel composition of open Petri nets. In particular, we prove two classification theorems: one for open Petri nets and one for monically open Petri nets (i.e. open Petri nets whose interfaces are specified by monic maps). Our results can be summarized as follows. The additive invariants of open Petri nets are completely determined by their values on a particular class of single-transition Petri nets. However, for monically open Petri nets, the additive invariants are determined by their values on transitionless Petri nets and all single-transition Petri nets. Our results confirm a conjecture of John Baez (stated during the AMS' 2022 Mathematical Research Communities workshop).
△ Less
Submitted 22 February, 2024; v1 submitted 2 March, 2023;
originally announced March 2023.
-
Learning What and Where to Draw
Authors:
Scott Reed,
Zeynep Akata,
Santosh Mohan,
Samuel Tenka,
Bernt Schiele,
Honglak Lee
Abstract:
Generative Adversarial Networks (GANs) have recently demonstrated the capability to synthesize compelling real-world images, such as room interiors, album covers, manga, faces, birds, and flowers. While existing models can synthesize images based on global constraints such as a class label or caption, they do not provide control over pose or object location. We propose a new model, the Generative…
▽ More
Generative Adversarial Networks (GANs) have recently demonstrated the capability to synthesize compelling real-world images, such as room interiors, album covers, manga, faces, birds, and flowers. While existing models can synthesize images based on global constraints such as a class label or caption, they do not provide control over pose or object location. We propose a new model, the Generative Adversarial What-Where Network (GAWWN), that synthesizes images given instructions describing what content to draw in which location. We show high-quality 128 x 128 image synthesis on the Caltech-UCSD Birds dataset, conditioned on both informal text descriptions and also object location. Our system exposes control over both the bounding box around the bird and its constituent parts. By modeling the conditional distributions over part locations, our system also enables conditioning on arbitrary subsets of parts (e.g. only the beak and tail), yielding an efficient interface for picking part locations. We also show preliminary results on the more challenging domain of text- and location-controllable synthesis of images of human actions on the MPII Human Pose dataset.
△ Less
Submitted 7 October, 2016;
originally announced October 2016.
-
A Finiteness Property of Torus Invariants
Authors:
Stella Gastineau,
Samuel Tenka
Abstract:
In this paper the invariant subring $R_n$ of an algebraic torus $T=(\mathbb{C}^\times)^r$ acting on the multi-homogenous polynomial ring
$$S^{\boxtimes n}=\bigoplus_{d=0}^\infty (S^{(d)})^{\otimes n},$$ where $S^{(d)}$ is the $d$th graded piece of the polynomial ring $S=\mathbb{C}[x_1,\dots,x_k]$, is studied from the viewpoint of matrices whose entries sum to zero. Using these weight matrices we…
▽ More
In this paper the invariant subring $R_n$ of an algebraic torus $T=(\mathbb{C}^\times)^r$ acting on the multi-homogenous polynomial ring
$$S^{\boxtimes n}=\bigoplus_{d=0}^\infty (S^{(d)})^{\otimes n},$$ where $S^{(d)}$ is the $d$th graded piece of the polynomial ring $S=\mathbb{C}[x_1,\dots,x_k]$, is studied from the viewpoint of matrices whose entries sum to zero. Using these weight matrices we prove that there exists a $d_1$ such that for all positive integers $n$, the relations of the invariant subring $R_n$ are generated in multi-homogenous degree $\leq d_1$. Grant: 0943832
△ Less
Submitted 28 October, 2015;
originally announced October 2015.
