Research in machine learning and AI, now a key technology in practically every industry and company, is too large for anyone to read. This column, Perceptron (formerly Deep Science), aims to collect some of the most relevant recent discoveries and papers – especially in, but not limited to, artificial intelligence – and explain why they are important.
This week in AI, researchers discovered a method that could allow an adversary to track the movements of a remote-controlled robot even when the robot’s communications are encrypted end-to-end. The co-authors, who are from the University of Strathclyde in Glasgow, say that their study shows that applying network security best practices is not enough to prevent attacks on autonomous systems.
Remote control, or remote control, promises to allow an operator to guide one or several robots remotely in multiple environments. Boot includes Pollen Robotics, Beamand A turtle demonstrated the usefulness of remote-controlled robots in grocery stores, hospitals, and offices. Other companies develop remotely controlled robots for tasks like bomb disposal or surveying sites with heavy radiation.
But new research shows that remote control, even when deemed “secure,” carries a risk of surveillance vulnerability. Co-authors Strathclyde describe in a paper using neural networks to infer information about what activities a remotely controlled robot is performing. After collecting samples of TLS– protected traffic between the robot and the controller and conducting analysis, they found that the neural network was able to identify movements about 60% of the time and also reconstruct the “process” warehousing workflow” (e.g. picking packages) with “high accuracy”.
Image credits: Shah et al.
A little alarm at once is a new thing research from researchers at Google and the University of Michigan explored people’s relationship to AI-enabled systems in countries with weak legislation and “national optimism” toward AI. The work surveyed India-based “financially stressed” users of instant lending platforms targeting borrowers with credit as determined by risk modeling AI. ro. According to the co-authors, users have experienced indebtedness for the “benefit” of instant loans and the obligation to accept harsh terms, share too much sensitive data, and pay high fees.
The researchers suggest that these findings suggest a need for greater “algorithmic accountability,” particularly as it relates to AI in financial services. “We argue that accountability is shaped by power relationships between users and platforms, and encourage policymakers to be cautious in adopting a purely technical approach,” they write. drugs to promote algorithmic accountability,” they wrote. “Instead, we call for positioning interventions that enhance user self-determination, create meaningful transparency, reconfigure the designer-user relationship, and deliver critical reflection in practitioners on broader accountability.”
In a little rain research, a team of scientists at TU Dortmund, University of Rhine-Waal and LIACS Universiteit Leiden in the Netherlands has developed an algorithm that they say can “solve” the game Rocket League. Driven to find a less computationally demanding way to generate gaming AI, the team leveraged what they call a “sim-to-sim” transfer technique, which helps train real AI systems perform in-game missions like goalkeeping and attacking in a stripped-down, simplified version of Rocket League. (Rocket League is basically like indoor soccer, except cars instead of players in teams of three.)
Image credits: Pleines et al.
It’s not perfect, but the researchers’ Rocket League play system, managed to save nearly all shots when the opponent’s keeper. When attacking, the system successfully scored 75% of the shots – a respectable record.
Simulators for human movements are also evolving at speed. Meta’s human limb tracking and simulation work has obvious applications in its AR and VR products, but it could also be used more widely in robotics and embodied AI. Research released this week received a tip from none other than Mark Zuckerberg.
Skeleton and muscle group simulation in Myosuite.
MyoSuite simulates muscles and skeletons in 3D as they interact with objects and with themselves – this is important for staff to learn how to hold and manipulate properly without damaging or dropping them, and Time in the virtual world also provides realistic grasping and interaction. It is said to run thousands of times faster for certain tasks, allowing simulated learning processes to happen much faster. “We are going to open source these models so that researchers can use them to advance the field further,” said Zuck. And they did!
A lot of these simulations are agent or object based, but this project from MIT consider simulating an overall system of independent actors: self-driving cars. The idea is that if you have a large number of cars on the road, you can have them work together not only to avoid collisions but also to prevent idling and unnecessary stops at lights.
If you look closely, only the cars ahead are actually stopped.
As you can see in the animation above, a set of autonomous vehicles communicating using the v2v protocol can essentially stop all but the cars ahead from stopping by gradually slowing down. degrees behind each other, but not so much that they actually stop. This type of driving behavior may not seem like it saves a lot of gas or battery, but when you scale it up to thousands or millions of cars, it makes all the difference – and it can be a great ride, too. ride more comfortably. However, good luck as everyone approaches such a perfectly spaced intersection.
Switzerland is taking an in-depth look at itself – using 3D scanning technology. The country is creating a giant map using UAVs equipped with lidar and other tools, but there’s a hitch: drone movements (intentionally and unintentionally) bring Errors in the point map need to be fixed manually. Doesn’t matter if you just scan one building, but the whole country?
Fortunately, a team at EPFL is integrating the ML model directly into the lidar capture stack that can determine when an object has been scanned multiple times from different angles and use that information to align the map. point map into a single coherent grid. This news not particularly illuminating, but The article that accompanies it goes into more detail. An example of the resulting map can be seen in the video above.
Finally, in some surprising but very exciting news about AI, a team from the University of Zurich has design an algorithm to track animal behavior so zoologists don’t have to go through weeks of footage to find two examples of courtship dances. It’s a collaboration with Zurich Zoo, which makes sense when you consider the following: “Our method can recognize even subtle or rare behavioral changes in research animals research, such as signs of stress, anxiety or irritability,” said lab leader Mehmet Fatih Yanik.
So the tool could be used both for learning and monitoring behaviors in captivity, for the well-being of animals kept in zoos, and for other forms of animal research. They can use fewer subject animals and get more information in less time, with less work by 10th graders poring over video files late into the night. That sounds like a win-win situation to me.
Image credits: Ella Marushenko / ETH Zurich
Also, love the illustrations.
