Research on machine learning and AI, now a key technology in virtually every industry and business, is far too extensive for anyone to read. This column aims to collect some of the most relevant recent discoveries and articles – particularly in the field of, but not limited to, artificial intelligence – and explain why they matter.
This week, AI applications have been found in several unexpected niches due to its ability to sift through large amounts of data or make sensible predictions based on limited evidence.
We’ve seen machine learning models adopting large data sets in biotechnology and finance, but researchers from ETH Zurich and LMU Munich are applying similar techniques to the data generated by international development aid projects such as disaster relief and housing. The team has trained its model on millions of projects (up to $2.8 trillion in funding) over the past 20 years, a massive dataset too complex to analyze in detail manually.
“You can think of the process as an attempt to read an entire library and sort similar books into topic-specific shelves. Our algorithm takes into account 200 different dimensions to determine how similar these 3.2 million projects are – an impossible workload for a human being,” said study co-author Malte Toetzke.
Top-level trends suggest that spending on inclusion and diversity has increased, while spending on climate has surprisingly fallen in recent years. You can view the dataset and trends they analyzed here.
Another area that few people think about is the large number of machine parts and components produced by various industries with a huge clip. Some can be reused, others recycled, others must be disposed of responsibly – but there are too many for human specialists to peruse. German R&D company Fraunhofer has developed a machine learning model to identify parts so they can be used instead of going to the scrap yard.
The system relies on more than ordinary camera footage as parts may look the same but be very different, or be mechanically identical but visually different due to rust or wear. So each part is also weighed and scanned by 3D cameras, and metadata such as provenance is also included. The model then suggests what it thinks the part is so that the human inspecting it doesn’t have to start all over again. It is hoped that tens of thousands of parts will soon be saved and the processing of millions accelerated by using this AI-assisted identification method.
Physicists have found an interesting way to apply the qualities of ML to an age-old problem. Essentially, researchers are always looking for ways to show that the equations governing fluid dynamics (some of which, like Euler’s, date back to the 18th century) are incomplete — that they break at certain extreme values. With traditional calculation techniques this is difficult, but not impossible. But researchers from CIT and Hang Seng University in Hong Kong propose a new deep learning method to isolate likely cases of fluid dynamics singularities, while others apply the technique to the field in other ways. This Quanta article explains this interesting development quite well.
Another age-old concept getting an ML layer is kirigami, the art of paper cutting that many will be familiar with in the context of making paper snowflakes. The technique goes back centuries, especially in Japan and China, and can produce remarkably complex and flexible structures. Researchers at Argonne National Labs were inspired by the concept to theorize a 2D material that can hold electronics on a microscopic scale but also bend it easily.
The team had done tens of thousands of experiments with manual 1-6 cuts and used that data to train the model. They then used a Department of Energy supercomputer to run simulations down to the molecular level. Within seconds, it produced a 10-cut variation with 40 percent stretch, far beyond what the team expected or even attempted on its own.
“It discovered things we never told it to find out. It learned something like a human learns and used its knowledge to do something else,” says project leader Pankaj Rajak. The success has prompted them to explore the complexity and scope of the simulation.
Another interesting extrapolation, performed by a specially trained AI, has a computer vision model that reconstructs color data from infrared input. Normally, a camera capturing IR would know nothing about the color of an object in the visible spectrum. But this experiment found correlations between certain IR bands and visible bands, creating a model to convert images of human faces captured in IR into those approaching the visible spectrum.
It’s still just a proof of concept, but such spectrum flexibility could be a useful tool in science and photography.
Meanwhile, a new study co-authored by Google AI leader Jeff Dean challenges the idea that AI is an environmentally friendly business because of its high computing requirements. While some studies have found that training a large model like OpenAI’s GPT-3 can generate carbon emissions equivalent to a small neighborhood, the Google-affiliated study claims that “by following best practices” the can reduce carbon emissions from machine learning by up to 1000x.
The practices in question relate to the types of models used, the machines used to train models, “mechanization” (e.g., cloud computing versus on-premises computers), and “map” (choosing data center locations with the cleanest energy). According to the co-authors, selecting only “efficient” models can reduce computation by factors of 5 to 10, while using processors optimized for machine learning training, such as GPUs, can reduce the performance-per-Watt ratio by factors from 2 to 5.
Any line of research that suggests that the impact of AI on the environment can be reduced is indeed cause for celebration. It should be noted, however, that Google is not a neutral party. Many of the company’s products, from Google Maps to Google Search, rely on models that required a lot of energy to develop and use.
Mike Cook, a member of the open research group Knives and Paintbrushes, points out that — even if the survey’s estimates are correct — there’s just no good reason for a company not to scale in an energy-inefficient way when it has benefits. them. While academic groups may pay attention to metrics such as carbon impact, businesses aren’t incentivized in the same way — at least right now.
“The whole reason we’re having this conversation to start is that companies like Google and OpenAI basically had infinite funding and chose to use it to build models like GPT-3 and BERT at any cost because they knew that it’s an advantage for them,” Cook told TechCrunch via email. “Overall, I think the paper is saying some nice things and it’s great when we think about efficiency, but I don’t think the issue is technical — we We know for sure that these companies will get big when they have to, they won’t hold back, so to say this is now fixed forever just feels like an empty line.”
The final topic of this week isn’t really exactly about machine learning, but rather what could be a way to simulate the brain in a more direct way. EPFL bioinformatics researchers created a mathematical model for creating tons of unique yet accurate simulated neurons that could eventually be used to build digital twins of neuroanatomy.
“The findings already enable Blue Brain to build biologically detailed reconstructions and simulations of the mouse brain, by reconstructing brain regions for simulations that replicate the anatomical properties of neuronal morphologies and include region-specific anatomy,” said study researcher Lida Kanari.
Don’t expect SIM brains to make for better AIs — this is very much geared towards advances in neuroscience — but perhaps the insights from simulated neuronal networks could lead to fundamental improvements in understanding the processes AI aims to digitally imitate.
This post AI cuts, flows and goes green – TechCrunch was original published at “https://techcrunch.com/2022/04/16/deep-science-ai-cuts-flows-and-goes-green/”