Moderator: Marcus Dahllöf, Program Director, MIT Startup Exchange Panelists:
As autonomous systems move out of the research laboratory into operational environments, they require ever deeper connections to their surroundings. Traditional notions of full autonomy have led to “clockwork” approaches where robots must be isolated from their human surroundings. Instead, we need precise, robust relationships with people and infrastructure. This situated autonomy appears in driverless cars' dependence on human-built infrastructure, the need for new systems of unmanned traffic management in the air, and the increasing importance of collaborative robotics in factories. How can we best design such systems to inhabit and enhance the human world? In this talk, David Mindell sketches a number of these emerging scenarios, traces new technologies to address the problems they raise, and envisions new approaches to human and robotic interaction that helps people and robots work together safely and collaboratively.
Computational Imaging systems consist of two parts: the physical part where light propagates through free space or optical elements such as lenses, prisms, etc. finally forming a raw intensity image on the digital camera; and the computational part, where algorithms try to restore the image quality or extract other type of information from the raw intensity image data. Computational Imaging promises to solve the challenge of imaging objects that are too small, i.e. of size at about the wavelength of illumination or smaller; too far, i.e. with extremely low numerical aperture; too dark, i.e. at very low photon counts; or too foggy, i.e. when the light has to propagate through a strongly scattering medium before reaching the detector. In this talk I will discuss the emerging trend in computational imaging to train deep neural networks (DNNs) to attack the quad of challenging objects. In several imaging experiments carried out by our group, objects rendered “invisible” due to various adverse conditions such as extreme defocus, scatter, or very low photon counts were “revealed” after processing of the raw images by DNNs. The DNNs were trained from examples consisting of pairs of known objects and their corresponding raw images. The objects were drawn from databases of faces and natural images, with the brightness converted to phase through a liquid-crystal spatial phase modulator. After training, the DNNs were capable of recovering unknown, i.e. hitherto not presented during training, objects from the raw images and recovery was robust to disturbances in the optical system, such as additional defocus or various misalignments. This suggests that DNNs may form robust internal models of the physics of light propagation and detection and generalize priors from the training set.
Layer 2 is currently used as an umbrella term for all operations that are performed “off chain” and use blockchains to settle transactions. This is based on the work of Tadge Dryja, who is one of the authors of the Lightning Network paper, and he continues to lead the DCI’s research in this area. The Lightning Network is one of the first applications of payment channels, and we’re confident we’ll see more. Another application we’ve been working on involves smart contracts. In order to create useful smart contracts, we need oracles, data feeds that verify real-world occurrences and submit this information in a format that can be used in a blockchain.