Contemporary sensorimotor learning approaches typically start with an existing complex agent (e.g., a robotic arm), which they learn to control. In contrast, this paper investigates a modular co-evolution strategy: a collection of primitive agents learns to dynamically self-assemble into composite bodies while also learning to coordinate their behavior to control these bodies. Each primitive agent consists of a limb with a motor attached at one end. Limbs may choose to link up to form collectives. When a limb initiates a link-up action and there is another limb nearby, the latter is magnetically connected to the 'parent' limb's motor. This forms a new single agent, which may further link with other agents. In this way, complex morphologies can emerge, controlled by a policy whose architecture is in explicit correspondence with the morphology. We evaluate the performance of these 'dynamic' and 'modular' agents in simulated environments. We demonstrate better generalization to test-time changes both in the environment, as well as in the agent morphology, compared to static and monolithic baselines. Project videos and code are available at https://pathak22.github.io/modular-assemblies/

We study a non-linear statistical inverse learning problem, where we observe the noisy image of a quantity through a non-linear operator at some random design points. We consider the widely used Tikhonov regularization (or method of regularization, MOR) approach to reconstruct the estimator of the quantity for the non-linear ill-posed inverse problem. The estimator is defined as the minimizer of a Tikhonov functional, which is the sum of a data misfit term and a quadratic penalty term. We develop a theoretical analysis for the minimizer of the Tikhonov regularization scheme using the ansatz of reproducing kernel Hilbert spaces. We discuss optimal rates of convergence for the proposed scheme, uniformly over classes of admissible solutions, defined through appropriate source conditions.

We give nearly matching upper and lower bounds on the oracle complexity of finding $\epsilon$-stationary points ($\| \nabla F(x) \| \leq\epsilon$) in stochastic convex optimization. We jointly analyze the oracle complexity in both the local stochastic oracle model and the global oracle (or, statistical learning) model. This allows us to decompose the complexity of finding near-stationary points into optimization complexity and sample complexity, and reveals some surprising differences between the complexity of stochastic optimization versus learning. Notably, we show that in the global oracle/statistical learning model, only logarithmic dependence on smoothness is required to find a near-stationary point, whereas polynomial dependence on smoothness is necessary in the local stochastic oracle model. In other words, the separation in complexity between the two models can be exponential, and that the folklore understanding that smoothness is required to find stationary points is only weakly true for statistical learning. Our upper bounds are based on extensions of a recent "recursive regularization" technique proposed by Allen-Zhu (2018). We show how to extend the technique to achieve near-optimal rates, and in particular show how to leverage the extra information available in the global oracle model. Our algorithm for the global model can be implemented efficiently through finite sum methods, and suggests an interesting new computational-statistical tradeoff.

During 14 years of intrepid exploration across Mars it advanced human knowledge by confirming that water once flowed on the Red Planet - but NASA's Opportunity rover has analysed its last soil sample. The robot has been missing since the US space agency lost contact during a dust storm in June last year and was declared officially dead on Wednesday, ending one of the most fruitful missions in the history of space exploration. Unable to recharge its batteries, Opportunity left hundreds of messages from Earth unanswered over the months, and NASA said it made its last attempt at contact. "I declare the Opportunity mission as complete," Thomas Zurbuchen, associate administrator of NASA's Science Mission Directorate told a news conference at mission headquarters in Pasadena, California. The community of researchers and engineers involved in the programme were in mourning over the passing of the rover, known affectionately as Oppy. "Spent the evening at JPL as the last ever commands were sent to the Opportunity rover on #Mars," Tanya Harrison, director of Martian research at Arizona State University, tweeted after a stint at Pasadena's Jet Propulsion Laboratory.

Hackers are attempting to cash in on the record-breaking popularity of Apex Legends, according to researchers who uncovered a fake app posing as an official mobile version of the new video game. Cyber security firm Check Point discovered what it believes to be the first malicious app of the free-to-play battle royale game, which is currently not available as a download for Android or iOS devices. Since launching just over a week ago, Apex Legends has attracted more than 25 million players, prompting comparisons to the viral success of Fortnite. Both games feature similar objectives and gameplay, though Apex Legends is not distributed on as many gaming platforms. Unlike Fortnite, Apex Legends can only be played on a PC, or consoles like the PS4, Xbox One and Nintendo Switch. But the fake app calling itself Apex Legends is actually an adware downloader, researchers found, meaning the people behind it earn revenue through online advertisements shown to the victim.

Nasa's Opportunity rover is officially dead, the space agency has said, after it disappeared in a dust storm on Mars. Clearly emotional Nasa staff, standing in front of a life-sized replica of the rover, said they had not heard back from the rover and that the mission would come to an end. Scientists described the difficult process of saying goodbye to the rover, which they had nicknamed Oppy and described as being like a beloved member of the family. "I am standing here with a sense of deep appreciation and gratitude," said Nasa associate administrator Thomas Zurbuchen, before he announced that the Opportunity mission is now considered complete. The robot set a huge number of records as it travelled across the Martian surface, eventually travelling some 28 miles and lasting far longer than any other Mars lander.

Ecosia, a German startup with an internet search engine, today, has brought in enough revenues to enable it to plant 50 million trees. This equates to the removal of 2.5 million tonnes of Co2 from the atmosphere, according to the company. Ecosia has used the profits from advertisements on its search engine to plant trees in Kenya, Brazil, Indonesia, Spain, Tanzania, Madagascar, Colombia, Peru, Senegal, Burkina Faso, Haiti, Morocco, Ethiopia, Uganda, Ghana and Nicaragua. Ecosia has partnered with Bing, Microsoft's search engine, to get results for users, but receives a majority portion of any revenues. After covering its internal costs, everything left goes towards planting trees; Ecosia is a non-profit organization.

TOKYO--(BUSINESS WIRE)--Feb 12, 2019--Mitsubishi Electric Corporation (TOKYO:6503) announced six new technologies, outlined below, at its annual Research and Development Open House held today at its headquarters in Tokyo, Japan, including three artificial intelligence (AI) technologies using the company's proprietary Maisart * AI. * M itsubishi Electric's AI creates the S tate-of-the- ART in technology The company announced today that it has developed an artificial intelligence (AI) technology capable of fast stepwise learning using a simulator to efficiently complete motion learning in a relatively short time using the company's Maisart AI technology. The company announced today that it has developed a unique behavioral-analysis artificial intelligence (AI) using the company's Maisart AI technology. The company announced today that it has developed what is believed to be the world's smallest power unit for a two-motor hybrid electric vehicle (two inverters and one converter), measuring just 2.7 liters in volume, offering a world-leading 150 kVA/L power density. The company announced today that it has developed sensing technology for highly accurate detection of vehicle perimeters even in dense fog or heavy rain. The company announced today that it has developed a simulation technology for predicting energy consumption and comfort levels to operate net zero energy buildings (ZEBs).

The common narrative in the tech industry is that those companies with access to the most data are best positioned to win with machine learning. The apparent logic of this positive feedback loop is simple: more/better data better models more accurate predictions better product or service more customers more data, etc. In his December 2018 blog post, Benedict Evans argues that this is too simplistic. It is critical to consider what data, for what problems, for which use cases? Google, Facebook, or China will not "just win" because of some combination of existing data assets, scale, and permissiveness around user data.

Apple is preparing for its next big launch event, according to a new report – but there might not be new products. The company is rumoured to be working on a whole host of new updates: an improved iPad Mini, a second version of the AirPods, as well as new Macs. That had many to hope that the company would hold one of its regular March events, and start selling many of those long-expected products. Apple will be holding one of those March events at its campus in California on 25 March, according to a new report from BuzzFeed. But those products won't be the central focus, according to the same report. Instead, the event will be headlined by Apple's news subscription service, which has been the subject of intense discussions between media organisations and the company.