Living on the edge: Zero Knowledge Human-assisted Machine Learning with Git and Bitcoin Part 1
I recently had a problem while upgrading my truck to autonomous driving. Here I present a solution, which may have utility beyond my own quibbles. I’ll do my best to describe my issue, and the solution which involves running a decentralized, zero-knowledge system, built with open-source tools such as git and bitcoin.
The Problem
While I was designing and building and testing the computer that powers the “brain” for the autonomous driving, I encountered an issue with power. In my rush to test the ML algorithms and run training from the cameras I’d mounted around the truck, I did not give the proper consideration to power requirements for the “brain”. My 25 year old truck electrical system bit the dust after a short time.
My first instinct was to attempt to solve the problem with an approach that seems to be the dominant recommended approach; upgrade your power, i.e. add more batteries, etc. This will solve the current problem of providing enough power to the “brain”, but wait…
The brain computer has the primary task of being the first line of connection between the kinetic interactions of the vehicle, and the algorithms that power them. This means the brain should be a powerful machine with low latency and handle operations in parallel. Before I begin going down the “big battery” rabbit hole, I am exploring the possibility of offloading some of the tasks of the brain, to a “distributed computer brain”, which will potentially reduce my power requirements for the “brain”.
The Solution
I present a zero knowledge, distributed computer system that provides nano-second latency compute power at the edge. Edge machines, such as the “brain” for my truck, can now offload the tasks required for real-time machine learning to the distributed computer system, including tasks requiring human computation in a privacy-preserving secure manner, using bitcoin.
Recent advances in computer software and hardware — specifically in the fields of data science and unbounded computer systems and models — now present us the opportunity to implement solutions to challenges that effectively require zero overhead in terms of computer hardware and other related capital expenditures previously required for running applications. The key to unlocking this reality is our ability to effectively implement our programs and systems across distributed computer systems. In this article, I explore building an un-bounded system with the goal of training and implementing a privacy-preserving machine learning model on edge devices using open systems such as Git and Bitcoin.
Introduction to Deep Neural Networks
An artificial neural network is a construction inspired by biological neural networks. The network learns by adjusting the weights of the connections between its artificial neurons, based on the input it receives and the errors it generates during the learning process. This learning process is typically done using an algorithm, such as backpropagation, which adjusts the weights in order to minimize the error between the predicted output and the actual output. The goal of an artificial neural network is to learn to perform a particular task, such as recognizing patterns, making decisions, or translating languages, by analyzing examples and making predictions based on that analysis. For more details how neural networks work check out https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
AI using DNN and human computation
Artificial Intelligence (AI) in the form of “Machine Learning” has recently seen tremendous advances, due in part to the availability of large amounts of data and the development of more powerful computational resources. These advances have led to the development of machine learning algorithms that can perform a wide variety of tasks, ranging from image and speech recognition to language translation and playing games.
Some of the key challenges facing the field include improving the ability of artificial intelligence systems to generalize from limited data, and developing methods for interpreting and explaining the decisions made by these systems.
Human Computation is a term that refers to the use of human intelligence and expertise to perform tasks that are difficult or impossible for computers to perform alone. This can take a variety of forms, such as crowd-sourcing, in which a large group of people are recruited to perform a specific task, or citizen science, in which volunteers contribute their time and skills to scientific research projects.
There are also limitations to human computation. It can be time-consuming and expensive to recruit and manage a large group of people, and the quality of the results can vary depending on the skill and motivation of the participants.
AI on Bitcoin
Bitcoin introduces a few interesting features that are attractive to computer developers and scientists looking to implement AI systems at scale. Some of those include the decentralized nature of the Bitcoin network, which allows for distributed processing and storage of data. This decentralization can make it more difficult for any single entity to control or manipulate the data, which can be beneficial in situations where trust is an issue.
Additionally, the use of blockchain technology in Bitcoin allows for a transparent and secure record-keeping system. This can be useful for tracking and verifying the results of AI algorithms, as well as for ensuring the integrity of the data being used to train those algorithms.
Finally, the use of Bitcoin’s native cryptocurrency, Bitcoin, as a means of payment can make it easier to incentivize participation in AI projects and to compensate those who contribute their time and resources to them.
Overall, the combination of decentralization, transparency, and a built-in payment system make Bitcoin an interesting platform for the development and deployment of AI systems at scale. For example, this implementation of handwritten digit classification is implemented completely in native bitcoin script!
In part 2, we dive into the world of Zero Knowledge Proofs on Bitcoin, Git, Linux, and a practical use case for quickly building and running scalable AI implementations with zero overhead.
References
AI on Bicoin — https://xiaohuiliu.medium.com/ai-on-bitcoin-96bbc97a62b9
Moodswings: A collaborative Game for Music Mood Label Collection — http://www.met-lab.org/emotion-recognition
