Jan 12, 2022
Segna Newsletter — 13 January 2022
What we’ve been reading
Cow noses are the fingerprint unlocking facial recognition technology for the livestock industry
ABC News
Researchers at the University of New England have discovered that the muzzle (nose) of a cow can be used to visually identify it with AI. Each cow’s muzzle contains a unique pattern, allowing the AI model to identify cows with over 99.1% accuracy. In addition, the technology is not limited to any specific breed of cattle but can be used for various breeds. The researchers highlight that “cattle are currently identified through some traditional methods such as ear tagging, tattooing, and so on, and all of these methods are not totally reliable.” This new approach is touchless, and “it doesn’t put any stress on the animals and it’s good for both the animals and humans.”
2021 was the year of monster AI models
MIT Tech Review
When GPT-3 was released in 2020, its scale at 175 billion parameters was unprecedented. However, it kicked off a new trend in ever-larger language models.
2021 saw the launch of AI21 Labs’s Jurassic-1 edging out GPT-3 with 178 billion parameters, followed by Gopher, a model released by DeepMind in December, which has 280 billion parameters. Megatron-Turing NLG (developed by Nivida and Microsoft) has 530 billion. Finally, Google’s Switch-Transformer and GLaM models have 1 and 1.2 trillion parameters, respectively.
The trend is not just in the US. For example, China saw the release of at least four massive language models; PanGu, Yuan 1.0, PCL-BAIDU Wenxin, and Wu Dao 2.0 at 200 billion, 245 billion, 280 billion, and 1.75 trillion parameters, respectively. Meanwhile, South Korea’s Naver announced a model called HyperCLOVA, with 204 billion parameters.
New machine learning algorithm could help in diagnosing MS sooner
Multiple Sclerosis News Today
A new machine-learning algorithm designed to analyze healthcare records could help diagnose Multiple Sclerosis (MS) sooner by identifying patients’ symptoms earlier. Symptoms of MS can appear years before a formal diagnosis, however, one of the greatest obstacles in diagnosing is that many of these early symptoms are mild and not specific to multiple sclerosis.
As an earlier diagnosis would allow for treatment to start sooner and thus lead to better outcomes, there has been a push to find new and more accurate ways to identify patients with early MS. The researchers behind the model are confident that “underlying biological signals must be present months or even years before diagnosis,” and that their solution “represents a new kind of ‘clear box’ explainable predictable models with broad applicability to other chronic medical conditions where early diagnosis can benefit patients.”
A cartel of influential datasets is dominating machine learning research new study suggests
Unite.ai
A new paper from the University of California and Google Research has found that 50% of all ‘benchmark’ machine learning datasets (like ImageNet) used were created by only 12 institutions — Stanford University, Microsoft, Princeton, Facebook, Google, the Max Plank Institute, AT&T, and a few more. While much of this is due to the prohibitive cost of creating one’s own dataset, one of the risks associated with this is overfitting — systems trained on these ‘gold standard’ datasets are unlikely to perform anywhere near as well on real-world data or even on new academic or original datasets.
It’s of-fish-ial: new method can identify specific fish calls
Koin
Using 18,000 hours of acoustic data collected from underwater microphones, researchers have developed a system that can accurately identify calls of damselfish. This has huge implications for monitoring ecosystems, providing an efficient and inexpensive way to understand changes in the marine environment due to climate change and other human-caused influences. “The benefit to observing fish calls over a long period of time is that we can start to understand how it’s related to changing ocean conditions, which influence our nation’s living marine resources. For example, damselfish call abundance can be an indicator of coral reef health.”
If you’d like to receive this newsletter in your inbox, you can subscribe here.
