By Nanistya Probosutedjo, Data Analyst at Zero One Technology
Welcome to our newsletter and glad to have you back for those who joined us last time!
First timers, feel free to head over to our first newsletter where we looked at the overarching theme of COVID-19, more specifically — pandemic misinformation detection on Facebook, contact tracing vs. exposure notifications, and wellness apps to help us cope during this difficult season.
The topic for today’s newsletter will focus on the agile methodology, its application in a deep learning model, and in Audi’s new Artemis unit. If you’re unfamiliar with the agile methodology, you can read more about it at the Agile Alliance website.
Improving accuracy of deep learning algorithms by mimicking the way children learn
Deep learning methods like neural networks were designed to imitate the way humans think. However, there’s a tendency to view neural networks as mature human brains. A recent study by Carnegie Mellon (CMU) introduced a methodology wherein neural networks were tasked to approach classification problems the way children do. The MIT Technology Review article depicts this by using the example of animal classification. When given images of animals, children are expected to distinguish between cats and dogs first before moving on to categorise animals as German Shepherds or Yorkshire Terriers. Following this methodology, the model was able to achieve up to 7% more accuracy.
The stepwise granularisation of categories had been executed in a previous 1990s study, deemed curriculum learning. In this previous study, datasets are divided for different levels of granularity or iteration. Hence, larger datasets are needed to ensure enough information is available for each learning phase. In contrast, the new CMU approach reuses the same dataset throughout the study, making it more data efficient.
What makes this specific neural network approach agile is its ability to maintain leanness and the implementation of iterations at each learning stage. As a result, companies and institutions are able to learn from less data at higher accuracy rates. This aligns with agile’s initial purpose of enabling successful response to change, especially in uncertain and turbulent times, much like the circumstances we are currently facing.
We encourage you to try this method on the datasets mentioned in our previous newsletter post, or other available datasets online!
Audi launches Artemis, a high-tech Electronic Vehicle (EV) unit planned to run on agile practices
This particular release is of interest to our discussion as we see agile practices transcend IT or software industrs in particular. What’s even more interesting is the fact that the agile methodology was birthed outside of IT. Hence, we see the return of agile practices to its original roots, or the growing adoption of it. Read more about the history of agile practises in this Harvard Business Review article.
Back to Artemis, this new unit serves as a testing ground for agile practices for the entire VW group. The initiative is expected to speed up the design to production life cycle process. Traditional methods have led car manufacturers to consume up to seven years for new models. However, Artemis is to reduce the length of this process by three years, and produce a “highly efficient electric car” by 2024. This comes with Audi’s plans to release 20 electric and 10 hybrid vehicles by 2025, and VW’s overall plan to release 75 electric vehicles across brands.
Although introduced as a competitive tool, the implementation of agile practices comes with its challenges. Audi AG CEO, Markus Duesmann, states that the company is working to implement the new approach without jeopardising manageability. It might be more difficult for larger companies who have been established for longer periods to transform in such a way, or experience general large-scale change.
Startups or smaller units looking to start using the agile methodology can start by looking at the resources provided in this Agile Alliance page.
You can also follow Zero One’s social media pages to see how we work with the agile mindset.
That concludes our second newsletter and we hope you’ll join us in our next one! We’d love to hear about your experience as new or experienced agile teams, so please feel free to respond and comment to the topics above.
Stay safe, well, and healthy!