Neural network can read tree heights from satellite images
Apr 2022, phys.org
We need to be reminded of how powerful data can be when it's f**king massive. It doesn't even have to be related. Like for example I can tell which zip code you grew up in, even the street and maybe even the house, based on really really fine-grained data about your teethbrushing habits, delivered by your smart toothbrush of course.
It sounds crazy, but given enough data, nothing is crazy.
We don't even have to know how to do it, just give a neural network enough data, and it will figure out the problem for you:
"Since we don't know which patterns the computer needs to look out for to estimate height, we let it learn the best image filters itself."
All you need are some training data, so in this case that means a bunch of trees for which we do know the height. Then we take the (otherwise flat) satellite data, mash it with the height data for the known trees to teach the network, and then unleash the network on the unknowns.
And why do we want to know how tall trees are? "Because whenever we cut down trees, we release carbon into the atmosphere, and we don't know how much carbon we are releasing."
via ETH Zurich: Nico Lang, Walter Jetz, Konrad Schindler, Jan Dirk Wegner, A high-resolution canopy height model of the Earth. arXiv:2204.08322v1 [cs.CV], arxiv.org/abs/2204.08322
Image source: The 4 Trends That Prevail on the Gartner Hype Cycle for AI, 2021, Gartner, Sep 2021
via: Intelligent Sensing: Enabling the Next “Automation Age” by Marco Cassis of STMicroelectronics at International Solid-State Circuits Conference 2022
Anxious individuals identified by analyzing their walking gait
May 2022, phys.org
The best method for identifying anxious individuals was walking. The team successfully identified people who were anxious with 75% accuracy.
They had to complete a balance test and a two-minute walk while wearing sensors. Based on these data, the team determined the young people who report being anxious walk in a way that's very similar to older adults who are fearful of falling. They find that young, anxious adults are constantly scanning for threats from side to side while walking and have trouble turning. The researchers also reported that anxious people have worse balance than those who are anxious.
via Clarkson University: Maggie Stark et al, Identifying Individuals Who Currently Report Feelings of Anxiety Using Walking Gait and Quiet Balance: An Exploratory Study Using Machine Learning, Sensors (2022). DOI: 10.3390/s22093163
Using electric signals from human brains, new software can perform computerized image editing
Jun 2022, phys.org
"All the existing software has been previously trained with labeled input. So, if you want an app which can make people look older, you feed it thousands of portraits and tell the computer which ones are young, and which are old.
Here, the brain activity of the subjects was the only input.
This is an entirely new paradigm in artificial intelligence—using the human brain directly as the source of input."
"All the existing software has been previously trained with labeled input. So, if you want an app which can make people look older, you feed it thousands of portraits and tell the computer which ones are young, and which are old. Here, the brain activity of the subjects was the only input. This is an entirely new paradigm in artificial intelligence—using the human brain directly as the source of input."
But alas:
"Collecting individual brain signals does involve ethical issues..."
We are the data, and we have lots of it.
Be a real shame if someone were to save our brainwaves and then sell them on the data market, where they could in turn unintentionally feed back to us our own biases after having been amplified by some behaviorally-exploitative algorithm...
via University of Copenhagen and University of Helsinki: Brain-Supervised Image Editing. Keith M. Davis III et al. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2022.
People with similar faces likely have similar DNA
Aug 2022, phys.org
They recruited human doubles from the photographic work of François Brunelle, a Canadian artist who has been obtaining worldwide pictures of look-alikes since 1999. They obtained headshot pictures of 32 look-alike couples.
Physical traits such as weight and height, as well as behavioral traits such as smoking and education, were correlated in look-alike pairs. Taken together, the results suggest that shared genetic variation not only relates to similar physical appearance, but may also influence common habits and behavior.
Eventually the all-seeing omnibot will be able to create a complete living human population gene network based only on our faces (and it will predict our habits and behaviors too).
via Barcelona Supercomputing Center and Josep Carreras Leukaemia Research Institute Barcelona: Manel Esteller, Look-alike humans identified by facial recognition algorithms show genetic similarities, Cell Reports (2022). DOI: 10.1016/j.celrep.2022.111257
Algorithm predicts crime a week in advance, but reveals bias in police response
Jul 2022, phys.org
Algorithm that forecasts crime by learning patterns in time and geographic locations from public data on violent and property crimes. The model can predict future crimes one week in advance with about 90% accuracy.
Not like that!
In a separate model, the research team also studied the police response to crime by analyzing the number of arrests following incidents and comparing those rates among neighborhoods with different socioeconomic status. They saw that crime in wealthier areas resulted in more arrests, while arrests in disadvantaged neighborhoods dropped. Crime in poor neighborhoods didn't lead to more arrests, however, suggesting bias in police response and enforcement.
via University of Chicago: Ishanu Chattopadhyay, Event-level prediction of urban crime reveals a signature of enforcement bias in US cities, Nature Human Behaviour (2022). DOI: 10.1038/s41562-022-01372-0
AI reveals unsuspected math underlying search for exoplanets
May 2022, phys.org
I forget about the singularity sometimes, but it's happening right now; we're already in the singularity:
Artificial intelligence (AI) algorithms trained on real astronomical observations now outperform astronomers in sifting through massive amounts of data to find new exploding stars, identify new types of galaxies and detect the mergers of massive stars, accelerating the rate of new discovery in the world's oldest science.
But that's already been the case, now it appears the AI has discovered "unsuspected connections hidden in the complex mathematics arising from general relativity".
The algorithm figured out new rules of gravitational microlensing:
"I argue that they constitute one of the first — if not the first — time that AI has been used to directly yield new theoretical insight in math and astronomy."-Joshua Bloom, UC Berkeley professor of astronomy
"Keming's machine learning algorithm uncovered this degeneracy that had been missed by experts in the field toiling with data for decades. This is suggestive of how research is going to go in the future when it is aided by machine learning, which is really exciting." -Scott Gaudi, professor of astronomy at Ohio State
via University of California - Berkeley: Keming Zhang et al, A ubiquitous unifying degeneracy in two-body microlensing systems, Nature Astronomy (2022). DOI: 10.1038/s41550-022-01671-6
Post Script:
"AI software has collaborated with mathematicians to successfully develop a theorem about the structure of knots, but the suggestions given by the code were so unintuitive that they were initially dismissed. Only later were they discovered to offer invaluable insight. The work suggests AI may reveal new areas of mathematics where large data sets make problems too complex to be comprehended by humans."
-DeepMind AI collaborates with humans on two mathematical breakthroughs, Matthew Sparkes, New Scientist, Dec 2021 [link]
They're talking about knots. Instead of feeding it facebook photos or product reviews, they gave it math problems...about knots. The "connection between algebraic and geometric invariants of knots".
And why knots? Because theories about knots can also be applied to quantum field theory. Don't ask me why, but they do, and it's called topology.
via Google DeepMind: Advancing mathematics by guiding human intuition with AI. Davies, A., Veličković, P., Buesing, L. et al. Nature 600, 70–74 (2021). DOI: 10.1038/s41586-021-04086-x
Post Post Script, On Synthetic Data:
'Fake' data helps robots learn the ropes faster
Jul 2022, phys.org
How to be a human:
For the rope-looping simulation and experiment, Mitrano and Berenson expanded the data set by extrapolating the position of the rope to other locations in a virtual version of a physical space—so long as the rope would behave the same way as it had in the initial instance. Using only the initial training data, the simulated robot hooked the rope around the engine block 48% of the time. After training on the augmented data set, the robot succeeded 70% of the time.
via University of Michigan: Data Augmentation for Manipulation, arXiv:2205.02886v3 [cs.RO] https://doi.org/10.48550/arXiv.2205.02886
