Causal Effects and Overlap in High-dimensional or Sequential Data

From Youtube: Large data sources such as electronic medical records or insurance claims present opportunities to study causal effects of interventions that are difficult to evaluate through experiments. One example is the management of septic patients in the ICU. This typically involves performing several interventions in sequence, the choice of one depending on the outcome of others. Successfully evaluating the effect of these choices depends on strong assumptions, such as having adjusted for all confounding variables. While many argue that having high-dimensional data increases the likelihood of this assumption being true, it also introduces new challenges: the more variables we use for estimating effects, the less likely that patients who received different treatments are similar in all of them. In this talk, we will discuss the role of overlap in causal effect estimation through the lens of domain adaptation and off-policy reinforcement learning.

Making intelligence intelligible - Rich Caruana

From Youtube: In the world of machine learning, there’s been a notable trade-off between accuracy and intelligibility. Either the models are accurate but difficult to make sense of, or easy to understand but prone to error. That’s why Dr. Rich Caruana, Principal Researcher at Microsoft Research, has spent a good part of his career working to make the simple more accurate and the accurate more intelligible. Today, Dr. Caruana talks about how the rise of deep neural networks has made understanding machine predictions more difficult for humans, and discusses an interesting class of smaller, more interpretable models that may help to make the black box nature of machine learning more transparent.

Fight or flight: the veterans at war with PTSD

From Youtube: One hundred years on from the end of the first world war, a group of veterans in Dorset are torn between their pride in their military careers and their anger over the lack of psychological support provided to them by the Ministry of Defence. With many feeling abandoned and left to battle significant mental health issues such as PTSD alone, former soldier Andy Price decides to take matters into his own hands, launching the Veteran’s Hub, a peer-to-peer support network for veterans and their families. Over the course of a year, the Guardian's Richard Sprenger follows Andy on his journey.

What if He Falls?

From Youtube: In 2017, when Alex Honnold made his stunning free-solo ascent of Yosemite’s El Capitan, he was taking an unimaginable risk: nearly three thousand feet of climbing without any ropes or safety equipment. But was the climb made even riskier by the filmmakers who accompanied him? In “What if He Falls?” filmmakers Elizabeth Chai Vasarhelyi and Jimmy Chin take us inside the process of documenting Honnold’s quest for climbing glory — and the ethical calculus of filming a friend who could, with the slip of a finger, plummet to his death.

This Brain Implant Could Change Lives

From Youtube: It sounds like science fiction: a device that can reconnect a paralyzed person’s brain to his or her body. But that’s exactly what the experimental NeuroLife system does. Developed by Battelle and Ohio State University, NeuroLife uses a brain implant, an algorithm and an electrode sleeve to give paralysis patients back control of their limbs. For Ian Burkhart, NeuroLife’s first test subject, the implications could be life-changing.

There's Waldo is a robot that finds Waldo

From Youtube: We built a little robot called "There's Waldo" to test the capabilities of Google's new AutoML Vision service. We've found that technologies can be unapproachable, and irrelevant by extension, to many people—so we learn ahead of the curve, and show our work in fun ways, to demonstrate what's possible. There's Waldo is a robot built to find Waldo and point at him. The robot arm is controlled by a Raspberry Pi using the PYUARM Python library for the UARM Metal. Once initialized the arm is instructed to extend and take a photo of the canvas below. It then uses OpenCV to find and extract faces from the photo. The faces are sent to the Google Auto ML Vision service which compares each one against the trained Waldo model. If a confident match of 95% (0.95) or higher is found the robot arm is instructed to extend to the coordinates of the matching face and point at it. If there are multiple Waldos in a photo it will point to each one. While only a prototype, the fastest There's Waldo has pointed out a match has been 4.45 seconds which is better than most 5 year olds.

60 Years of Challenges and Breakthroughs by DARPA

From Youtube: With a focus on the people and perseverance behind DARPA’s ability to make the impossible possible, trace the agency’s history from its charter following the Soviet Union’s Sputnik launch to advances across a spectrum of technologies. Building on a legacy of innovation, DARPA continues to push technological boundaries to ensure U.S. military superiority and serve the people who serve and protect our nation.

What People See in a Robot: A New Look at Human-Like Appearance

From Youtube: What People See in a Robot: A New Look at Human-Like Appearance. A long-standing question in HRI is what effects a robot’s human-like appearance has on various psychological responses. A substantial literature has demonstrated such effects on liking, trust, ascribed intelligence, and so on. Much of this work has relied on a construct of uni-dimensional low to high human-likeness. I introduce evidence for an alternative view according to which robot appearance must be described in a three-dimensional space, encompassing Body/Manipulators (e.g., torso, arms, legs), Facial Features (e.g., head, eyes), and Surface Look (e.g., eyelashes, skin, genderedness). The broad human-likeness concept can thus be decomposed into more concrete appearance dimensions, and robots’ degrees of human-likeness are constituted by different combinations of these dimensions. In a study using 24 robots selected from this three-dimensional appearance space, I then show that the different dimensions separately predict inferences people make about the robot’s affective, social-moral, and physical capacities.


Subscribe to Ricardo Martins RSS