Jan 26, 2022
Segna Newsletter — 27 January 2022
What we’ve been reading
Scientists are sequencing the genome of every complex species on Earth
Singularity Hub
The Earth Biogenome Project is a global consortium that aims to sequence the genomes of all complex life on Earth in 10 years. The project will involve sequencing 1.8 million described species across plants, animals, fungi, and eukaryotes. Launched in 2018, it is currently in phase one — sequencing one genome from each of the 9,400 taxonomic families on Earth.
The entire Earth Biogenome Project will cost less in today’s dollars than the Human Genome Project.
How well do explanation methods for machine-learning models work?
Tech Xplore
Imagine a neural network is used to detect cancer in mammogram images; while the model seems to be performing well, it’s focusing on features such as watermarks or timestamps rather than actual signs of tumors. How would we know?
MIT researchers have devised a process to test this by modifying data to evaluate whether neural networks can correctly identify important features. Their results found that popular methods often miss the important features in an image, and most barely reached a random baseline level of 50%, while some methods even performed worse than the baseline.
The biggest danger isn’t Skynet — it’s human bias that should scare you
Interesting Engineering
Artificial general intelligence (AGI), or a system that can understand or learn any intellectual task a human can, is often railed against as “the single biggest existential crisis that we face and the most pressing one.” However, few people rail against the ways that AI systems are already harming humans and the reasons why.
There’s a saying in statistics and data science: garbage in, garbage out. This is a serious problem when talking about machine learning algorithms that learn from historical data as they ultimately come to reflect, exacerbate, and automate injustices in the world already.
Classifying plastics with hyperspectral camera analysis and unsupervised machine learning
AZo Cleantech
To recycle plastic, different types of plastic must be sorted with at least 95% purity. However, current sorting methods — flotation or mechanical, often only sort with 75–95% purity. Researchers at Aarhus University applied machine learning on hyperspectral data to build a model for the classification of plastics. Their research found that using commercially available spectrographs and minimal data processing, they were able to distinguish 13 different plastics. This is promising as a tool for addressing recycled plastic purity and, thus, the global plastic challenge.
Machine learning algorithms help scientists explore Mars
EOS
Researchers are using unsupervised machine learning to analyze data collected from NASA’s Mars Curiosity Rover to understand the sedimentary structure of the Gale crater — where it landed. The machine learning system, which classifies spectra data into distinct geological regions based on chemical composition, may prove to be a powerful tool to map Mars’ surface on a large scale, particularly as more rovers explore Mars.
If you’d like to receive this newsletter in your inbox, you can subscribe here.
