System Bits: Nov. 17

Algorithmic photo captioning
Researchers at Idiap, an EPFL-affiliated research institute in Martigny have developed an algorithm that can describe an image without having to pull up captions that it has already learned by using a program that makes vector representations of images and captions based on an analysis of caption syntax.

Rémi Lebret, a PhD student specializing in Deep Learning at Idiap explained that when the program is given a photo, it compares the image vector to the vector of possible words and selects the most likely noun, verb and prepositional phrases. This is how the system finds the most likely description for a photo of a man skateboarding, for example, even if it has never seen a similar photo previously. The computer breaks down the picture into elements (“a skateboard, a man, a ramp”) and verbs that could describe the action (« riding») before captioning the picture.

(Source: EPFL)

Algorithm speeds up complex modeling
MIT researchers remind that to work with computational models is to work in a world of unknowns, and models that simulate complex physical processes are staggeringly complex, sometimes incorporating hundreds of parameters, each of which describes a piece of the larger process.

Now, however, these researchers have created an algorithm that vastly reduces the computation of virtually any computational model, and may be thought of as a shrinking bull’s-eye that, over several runs of a model, and in combination with some relevant data points, incrementally narrows in on its target: a probability distribution of values for each unknown parameter.
With this method, the MIT team said they were able to arrive at the same answer as a classic computational approaches, but 200 times faster.

Youssef Marzouk, an associate professor of aeronautics and astronautics at MIT, said the algorithm is versatile enough to apply to a wide range of computationally intensive problems — in a range of fields from engineering and geophysics to subsurface modeling, very often with unknown parameters.

He said they wanted to treat the model as a black box and find out if they could accelerate the process in some way — and that’s what the algorithm does.

With this algorithm, the researchers intend to significantly speed up the conventional sampling process by short-circuiting this model and put in an approximate model, which may be orders of magnitude cheaper to evaluate.

They said the algorithm can be applied to any complex model to quickly determine the probability distribution, or the most likely values, for an unknown parameter.

Pop-up sensors give flexible surgical robots a sense of touch
While more and more surgeries are performed from behind a computer console as multi-million dollar, multi-armed surgical robots like the Zeus or Da Vinci systems replace hand-held scalpels — there is still something human hands can do better than robotic arms: feel. Thus far, no robotic tool can match the human hand in its ability to sense and adjust force.

However, researchers at Harvard University have moved a step closer to that goal through the development of a new method to build low-cost, millimeter-scale force sensors.

They explained that biggest challenge in developing force sensors for robotic surgical tools is size. Soft robotic surgical systems, for obvious reasons, need to be small and the sensors that sit on the system’s robotic fingertips need to be even smaller. Current conventional fabrication techniques limit the complexity and the sophistication of these millimeter-sized sensors while significantly driving up the cost of assembly and implementation, which poses a barrier to widespread adoption of force-sensing soft robotic surgical tools that can perform minimally invasive and complex surgeries in an inherently safe way.

Image of the sensor encapsulated into the catheter with a US penny for scale, and (right) a photorealistic rendering of the molded system. (Source: Harvard)

To solve this manufacturing problem, the team turned to pop-up manufacturing. Inspired by origami and pop-up books, this technique fabricates complex micromachines by layering laser-cut materials into thin, flat plates that pop up into a complete electromechanical devices. This fabrication approach has the potential to remove the human element from the assembly process by allowing devices to ‘build themselves,’ significantly driving down the cost of manufacturing, the research team added.