Future implementations
First, the machine learning part of this project and the results of classifying names need to be improved. Perhaps the window of the RNN through which sequential patterns are shown needs to be bigger.
Moreover, the representation of names is done at the character-level, with one-hot encoding whose alphabet has 56 possible characters. This little amount of characters cannot represent languages with different characters like Chinese, Japanese, Hebrew, and Russian. A more comprehensive set of characters, perhaps a bigger subset of Unicode, can be used in future projects.
The current implementation of this project splits the dataset randomly. However, it is much better to follow the philosophy of SPAIC: Each Raspberry Pi is a device connected to the Internet and each user can add their local names to the dataset. All the local names in local devices can be shared through Federated Learning with a trusted aggregator in order to create a more robust model. In that way, the privacy of such local contributions is preserved and at the same time, the machine learning model is enriched with new names from other parts of the world. This Federated Learning process is done like in the following graph:
Federated Learning - Image taken from https://www.slideshare.net/MindosCheng/federated-learning
If Raspberry Pi is a general computer capable of running Linux, perhaps there is a lightweight version of Docker https://www.docker.com/ that can be installed in Raspberry Pi. This could simplify a lot the process of installing all the software required in a Raspberry Pi. Because we just create a docker script to download everything and put it inside Raspberry Pi. So, the process will be super simple. Once the docker script runs well in one Raspberry Pi, it is guaranteed that the docker script will run well in almost all Raspberry Pis.
The advent of federated learning has resulted in a brand new area in artificial intelligence (AI) research. These days, consumer devices including mobile phones daily produce enormous amounts of data that carry the user’s personal information and preferences such as their most visited websites, their social media app use, and their most-watched video types. These data carrying valuable user information, are used to improve and personalize the device to power the device experience based on the individual’s usage. What makes federated learning unique is its ability to mask raw data (hence reserving user’s privacy) while helping train a better model to enhance the service for the user.
Federated learning is reshaping how ML models are prepared. Google have recently debuted their first federated learning program that is predicted by Professor Mi Zhang (Professor of Electrical and Computer Engineering and Computer Science) give rise to several applications including next-word prediction, on-device ranking, and content suggestion. Subsequently, ML models can be trained in the absence of computer resources belonging to AI organizations and the user’s privacy will no longer be shared in return for enhanced services.
Federated learning has become an attractive field of research in industrial applications such as health care. It is boosting a medical platform (from Owkin), which assists medical professionals to perform tests and experiment for prediction of disease evolution drug toxicity. Recently, AI researchers from MIT, Harvard University, CSAIL and Tsinghua University’s Academy of Arts and Design devised a technique to analyze electronic medical records using federated learning.
All in all, federated learning and approaches which deliver machine intelligence without interfering with personal data is likely to gain popularity as everyone is increasingly concerned about privacy and more device producers turn to on-device machine learning.
Resources
https://towardsdatascience.com/the-new-dawn-of-ai-federated-learning-8ccd9ed7fc3a
As the Internet of Things (IoT) is well on its way to grow to 25 billion devices in 2020 (according to Gartner and IDC), the world is entering an age where technology will become smaller, more affordable and way more inconspicuous. The Raspberry Pi (RPi) is a strong example of this era. Following its launch in the University of Cambridge as a board for DIY projects in 2011, the RPi swiftly gained huge popularity and 12 million units were sold. The RPi’s success has two important implications:
-
“It is a harbinger for the prowess of the Internet of Things” as said by Jason Hiner
-
It demonstrates that extremely low-cost devices can be adequate for many users and several tasks
It has become well-known as one of the most flexible and affordable computers of our time. Following its eruption, the RPi has been a catalyst for engineers, IT experts, makers, and startups to debut all types of new experiments. The RPi’s modest cost makes it a safety net for small startups which is a huge advantage to aspiring and budding entrepreneurs. It can be used to begin small-scale projects requiring low computational power with just a few clicks. Additionally, the RPi is widely used to prototype new ideas by teams of tech enthusiasts in hackathons. It is used to power robots, drones and weather stations, and in-home automation, building a mini web browser or living room PC, all applications unimaginable by its creator Eben Upton during its launch. Furthermore, some teams have accomplished converting their RPi into a low-cost personal computer with a function highly comparable to a $1000 computer. Due to its low-cost and power consumption combined with relatively powerful computing and multimedia functionality, the RPi is also gaining attention from developing countries where computer equipment is not always readily available and electricity is expensive. Computing at a low-cost for everyone would interest and educate more people into computer science globally which subsequently would bring about in useful innovation and higher access to technological advances in regions where resources are limited.
Hence, the RPi’s recruitment rate holds huge future promises where more computational experimentations are expected to be performed with it. Today, users are increasingly settling for low-cost devices which sufficiently serve their needs, and the RPi is at the forefront of this trend.
https://www.zdnet.com/article/why-raspberry-pi-is-the-future-of-computing-devices/
https://thenextweb.com/dd/2019/02/07/raspberry-pis-eben-upton-opens-up-about-the-future-of-retail/