arXiv subjects

Michael Bukatin

Publications and source records attributed to Michael Bukatin.

12 recordsLinked to original sources

arXiv2023

Safety without alignment

Currently, the dominant paradigm in AI safety is alignment with human values. Here we describe progress on developing an alternative approach to safety, based on ethical rationalism (Gewirth:1978), and propose an inherently safe implementation path via hybrid theorem provers in a sandbox. As AGIs evolve, their alignment may fade, but their rationality can only increase (otherwise more rational ones will have a significant evolutionary advantage) so an approach that ties their ethics to their rationality has clear long-term advantages.

cs.AI↗

arXiv2018

Notes on Pure Dataflow Matrix Machines: Programming with Self-referential Matrix Transformations

Dataflow matrix machines are self-referential generalized recurrent neural nets. The self-referential mechanism is provided via a stream of matrices defining the connectivity and weights of the network in question. A natural question is: what should play the role of untyped lambda-calculus for this programming architecture? The proposed answer is a discipline of programming with only one kind of streams, namely the streams of appropriately shaped matrices. This yields Pure Dataflow Matrix Machines which are networks of transformers of streams of matrices capable of defining a pure dataflow matrix machine.

cs.PL↗

arXiv2018

Programming Patterns in Dataflow Matrix Machines and Generalized Recurrent Neural Nets

Dataflow matrix machines arise naturally in the context of synchronous dataflow programming with linear streams. They can be viewed as a rather powerful generalization of recurrent neural networks. Similarly to recurrent neural networks, large classes of dataflow matrix machines are described by matrices of numbers, and therefore dataflow matrix machines can be synthesized by computing their matrices. At the same time, the evidence is fairly strong that dataflow matrix machines have sufficient expressive power to be a convenient general-purpose programming platform. Because of the network nature of this platform, programming patterns often correspond to patterns of connectivity in the generalized recurrent neural networks understood as programs. This paper explores a variety of such programming patterns.

cs.PL↗

arXiv2018

Dataflow matrix machines as programmable, dynamically expandable, self-referential generalized recurrent neural networks

Dataflow matrix machines are a powerful generalization of recurrent neural networks. They work with multiple types of linear streams and multiple types of neurons, including higher-order neurons which dynamically update the matrix describing weights and topology of the network in question while the network is running. It seems that the power of dataflow matrix machines is sufficient for them to be a convenient general purpose programming platform. This paper explores a number of useful programming idioms and constructions arising in this context.

cs.NE↗

arXiv2018

Dataflow Matrix Machines as a Generalization of Recurrent Neural Networks

Dataflow matrix machines are a powerful generalization of recurrent neural networks. They work with multiple types of arbitrary linear streams, multiple types of powerful neurons, and allow to incorporate higher-order constructions. We expect them to be useful in machine learning and probabilistic programming, and in the synthesis of dynamic systems and of deterministic and probabilistic programs.

cs.NE↗

arXiv2018

Dataflow Matrix Machines and V-values: a Bridge between Programs and Neural Nets

1) Dataflow matrix machines (DMMs) generalize neural nets by replacing streams of numbers with linear streams (streams supporting linear combinations), allowing arbitrary input and output arities for activation functions, countable-sized networks with finite dynamically changeable active part capable of unbounded growth, and a very expressive self-referential mechanism. 2) DMMs are suitable for general-purpose programming, while retaining the key property of recurrent neural networks: programs are expressed via matrices of real numbers, and continuous changes to those matrices produce arbitrarily small variations in the associated programs. 3) Spaces of V-values (vector-like elements based on nested maps) are particularly useful, enabling DMMs with variadic activation functions and conveniently representing conventional data structures.

cs.NE↗

arXiv2017

Revisiting EPRL: All Finite-Dimensional Solutions by Naimark's Fundamental Theorem

In this paper we research all possible finite-dimensional representations and corresponding values of the Barbero-Immirzi parameter contained in EPRL simplicity constraints by using Naimark's fundamental theorem of the Lorentz group representation theory. It turns out that for each non-zero pure imaginary with rational modulus value of the Barbero-Immirzi parameter $γ= i \frac{p}{q}, p, q \in Z, p, q \ne 0$, there is a solution of the simplicity constraints, such that the corresponding Lorentz representation is finite dimensional. The converse is also true - for each finite-dimensional Lorentz representation solution of the simplicity constraints $(n, ρ)$, the associated Barbero-Immirzi parameter is non-zero pure imaginary with rational modulus, $γ= i \frac{p}{q}, p, q \in Z, p, q \ne 0$. We solve the simplicity constraints with respect to the Barbero-Immirzi parameter and then use Naimark's fundamental theorem of the Lorentz group representations to find all finite-dimensional representations contained in the solutions.

gr-qc↗

arXiv2017

Barbero-Immirzi parameter as a solution of the simplicity constraints

In this paper we naturally obtain the values of the Barbero-Immirzi parameter as the solution of the simplicity constraints rather than setting it a priori. Particularly the Main theorem shows that if $γ= \pm i$ then the simplicity constraints require that the corresponding Lorentz group representations be necessary finite dimensional and therefore non-unitary.

gr-qc↗

arXiv2017

Dataflow Matrix Machines as a Model of Computations with Linear Streams

We overview dataflow matrix machines as a Turing complete generalization of recurrent neural networks and as a programming platform. We describe vector space of finite prefix trees with numerical leaves which allows us to combine expressive power of dataflow matrix machines with simplicity of traditional recurrent neural networks.

cs.NE↗

arXiv2016

Dataflow Graphs as Matrices and Programming with Higher-order Matrix Elements

We consider dataflow architecture for two classes of computations which admit taking linear combinations of execution runs: probabilistic sampling and generalized animation. We improve the earlier technique of almost continuous program transformations by adopting a discipline of bipartite graphs linking nodes obtained via general transformations and nodes obtained via linear transformations which makes it possible to develop and evolve dataflow programs over these classes of computations by continuous program transformations. The use of bipartite graphs allows us to represent the dataflow programs from this class as matrices of real numbers and evolve and modify programs by continuous change of these numbers. We develop a formalism for higher-order dataflow programming for this class of dataflow graphs based on the higher-order matrix elements. Some of our software experiments are briefly discussed.

cs.PL↗

arXiv2016

Almost Continuous Transformations of Software and Higher-order Dataflow Programming

We consider two classes of stream-based computations which admit taking linear combinations of execution runs: probabilistic sampling and generalized animation. The dataflow architecture is a natural platform for programming with streams. The presence of linear combinations allows us to introduce the notion of almost continuous transformation of dataflow graphs. We introduce a new approach to higher-order dataflow programming: a dynamic dataflow program is a stream of dataflow graphs evolving by almost continuous transformations. A dynamic dataflow program would typically run while it evolves. We introduce Fluid, an experimental open source system for programming with dataflow graphs and almost continuous transformations.

cs.PL↗

arXiv2015

Linear Models of Computation and Program Learning

We consider two classes of computations which admit taking linear combinations of execution runs: probabilistic sampling and generalized animation. We argue that the task of program learning should be more tractable for these architectures than for conventional deterministic programs. We look at the recent advances in the "sampling the samplers" paradigm in higher-order probabilistic programming. We also discuss connections between partial inconsistency, non-monotonic inference, and vector semantics.

cs.LO↗