Before we get started with this chapter, here is the full summary video, containing all 5 previous parts, enjoy!
The first 5 chapters of this blog post series summarize the mitigation part of the paper, which is concerned with preventing or slowing down climate change. The paper however contains another major part called adaption, which is not covered in this series. This part discusses opportunities how humanity can use machine learning to better adapt to upcoming changes and crisis. Another small part of the paper discusses meta tools, among other things how ML can improve personalized education. I did not have enough time (yet) to produce summaries of these parts. The reason they are not included does not mean that they are less relevant. So while you check out datasets and further resources, you might want to have a look at these aspects as well.
This is part six of a six-part series:
- Electricity Systems
- Buildings & Cities
- Farms & Forests
- Industry & Carbon Dioxide Removal
- Datasets & further resources
Datasets & further resources
This chapter contains the following sections:
The energy impact of machine learning
Before you start to explore further resources and train models, let’s have a quick look at the energy impact of machine learning itself. Training complex machine learning models might consume amounts of energy one would not expect. One case is posed by researchers, who estimate that a NLP deep learning training pipeline with tuning and experimentation can cause more than double of the emissions an average American citizen causes over one year of time. Researchers suggest to prioritize computationally efficient hardware and algorithms. The table below makes some easy to grasp comparisons:
Estimated carbon footprint from training common NLP models in comparison with other causes of emission [Strubell et al., 2019]
Solutions often depend on policies
Machine Learning is not a miracle cure and cannot solve all climate change related problems. The researchers point out that ML is an invaluable tool but policy makers must decide to act to drive large-scale progress. For some solutions there is an economic incentive to implement them, such as the recognition of leaks in gas pipelines where lost resources mean lost profit as well. For many other solutions, there need to be policies that give companies and citizens incentives to opt for low-carbon decisions. Though, as we learned in the last chapters, there are solutions where ML can enable better policy decisions, for example when city planners can predict transportation demand more accurately and as a result create better public transit.
There are publicly available datasets related to climate change at climatechange.ai/resources categorized by the chapters of the paper. Datasets range from satellite sensors about methane to the energy intensity of buildings in New York City. NASA also provides terabytes of data from satellite sensors through their EARTHDATA initiative. The Earth Engine Data Catalog from Google also collects various datasets consisting of satellite data. One has to request access to the datasets which are made available for personal use.
If you stumble open other datasets which should be mentioned here, please contact me or comment below.
I hope that this series inspired you to dig deeper into this matter. If so, i highly recommend to check out the climate change talks at ICML 2019 where researchers give talks with more depth into the topics of this series. Two of many more talks are embedded in the following:
John Platt (Google AI): AI for Climate Change – the context
Andrew NG (Stanford): Tackling climate change challenges with AI through collaboration
Also check out Project Drawdown, where climate change solutions are collected and ranked by their importance.
To learn more about what my colleagues do with machine learning, have a look at our other ML related blog posts.
Get in touch
I am looking forward to get to know like minded people. Feel free to get in touch.
Many thanks to all researchers of the paper:
David Rolnick, Andrew Y. Ng, Priya L. Donti, Lynn H. Kaack, Kelly Kochanski, Alexandre Lacoste, Kris Sankaran, Andrew Slavin Ross, Nikola Milojevic-Dupont, Natasha Jaques, Anna Waldman-Brown, Yoshua Bengio, Alexandra Luccioni, Tegan Maharaj, Evan D. Sherwin, S. Karthik Mukkavilli, Konrad P. Kording, Carla Gomes, Demis Hassabis, John C. Platt, Felix Creutzig and Jennifer Chayes.