I’ve wanted to do this work for years! It’s a simple idea on how to get below the Shannon limit by reordering data into a more compressible data source, you lose order but gain compression ratio. This works well on data that is usually difficult to compress and is open source here’s the repo it contains the source to my ISIT25 poster as well. ENJOY!!

Abstract 🔗We demonstrate a many operand asynchronous dot product with adjustable bit width for one of the multiplication tuple operands. The generalised algorithm called MADD relies on the commutative aspects of both the addition order and multiplication tuple operands. The approach is deeply wedded to asynchrony due to optimisation of the variable bitwidth memory architecture. Although specialised, this algorithm has a significant practical application in deep learning models particularly in managing the variable operand size due to the data variability, layering and optimisation process.

A couple more videos, a full length description including the MADD algorithm and next steps in opensourcing key elements. There should have been a video here but your browser does not seem to support it. There should have been a video here but your browser does not seem to support it.

The first introductory video, hope you enjoy! There should have been a video here but your browser does not seem to support it.

A New Approach to Vector Multiplication and Matrix Operations using Temporal Computing 🔗Introduction 🔗The technology we discuss in this concept paper relates to a radical rethink of both computer architecture and the method of arithmetic used for calculation. Our target operation is the vector dot product $\sum_{i=0}^Iw_ix_i$ , which is the build block of matrix multiplication, which is the building block of all AI.. This note’s thesis is that to make a radically more efficient operator we need to fundamentally rethink both the data representation and the underlying mode of storage and operation. Temporal processing fits nicely with both of these, as it utilises unary codes, and manipulates and operates on this representation using time-delays for storage. So far we have built this in c-mos (using positions in arrays, which we call “index space”) and require funding to move this to a “fully” temporal instantiation.

Taoism notes 🔗Introduction 🔗Taoism is a practical self consistent personal philosophy that despite its ancient roots fits nicely with modern scientific thought. This including concepts such as the block universe, recursive structures, quantum theory, fuzzy logic and non-monotonic reasoning. It provides best in class guidance for several practical areas including how to generally conduct oneself, how to interact with others, and quite brilliantly how to think about the universe holistically. No stone is left unturned by the broadness of it’s reach. In addition, and for free, it describes the well in which the water of creativity is drawn.

This is a bit different to the usual technical stuff I write. I want to talk about LLMs. Whilst an amazing piece of technology (magic), they do play heavily into the hands of large corporations. I suppose this stuff is quite obvious but I thought I’d write it anyhow! Why is this? They commodify skill and knowledge such that it is centralised, and no longer distributed amongst a variety of places and humans, like the internet. Whilst this initially seems like a great idea, once this is behind a paywall we will all be reliant on it in the extreme, and ultimately knowledge disparity will occur in all it’s forms.

This short post is a call to arms for Tsetlin Machines (TMs), a somewhat academic outsider’s view of what the community has and what is missing. The discourse is biased to commercial usages of the technology, a vital step forward in establishing the true value of the algorithms discussed in this blog post.. Introduction 🔗Tsetlin Machines are a new class of classifier that have significant usability and performance benefits over both standard classifiers, and most interestingly over existing deep learning (DL) systems. Clearly to establish the usefulness of the algorithm the technology needs to step outside of the academic community and become accepted in a wider context, which inevitably means touching commercial systems. In this paper we attempt to describe steps that will help make this possible.

Hi All! I recently delivered a talk about polars at PyDataLondon23. Here’s the slides data and notebook. Hope you find them useful! This data has ben created as an open data set by CitizenMe as part of their 360° Data Lab initiative.

The ability to perform accurate repetitive computation has been central to a large number of scientific and technological advances in the last seventy years. At the heart of this commercially is Moore’s Law, which states that computational power provided by traditional computer central processing units will double year-on-year for the foreseeable future. Unfortunately, several factors have compounded to make this less likely to continue. Heat dissipation, atomic and quantum effects provide practical limits to the miniaturisation and packing of transistors, and the limited bandwidth between CPU and memory limits computation speed. There is some doubt whether fabrication can extend below 3nm.