This paper addresses the lower limits of encoding and processing the information acquired through interactions between an internal system (robot algorithms or software) and an external system (robot body and its environment) in terms of action and observation histories. Both are modeled as transition systems. We want to know the weakest internal system that is sufficient for achieving passive (filtering) and active (planning) tasks. We introduce the notion of an information transition system for the internal system which is a transition system over a space of information states that reflect a robot's or other observer's perspective based on limited sensing, memory, computation, and actuation. An information transition system is viewed as a filter and a policy or plan is viewed as a function that labels the states of this information transition system. Regardless of whether internal systems are obtained by learning algorithms, planning algorithms, or human insight, we want to know the limits of feasibility for given robot hardware and tasks. We establish, in a general setting, that minimal information transition systems exist up to reasonable equivalence assumptions, and are unique under some general conditions. We then apply the theory to generate new insights into several problems, including optimal sensor fusion/filtering, solving basic planning tasks, and finding minimal representations for modeling a system given input-output relations.

Classical methods, such as ab initio calculations and molecular dynamics simulations, compute atomic and electronic states based on fundamental principles of quantum and classical mechanics (Figure 1A.). Although accurate, these methods are restricted to materials with simple structures, like molecules and crystals, and struggle with complex materials at larger scales. Advances in computational materials science have significantly broadened the range of materials that can be addressed, not only through improvements in classical approaches but also through new data-driven methods, including machine learning, deep learning, and generative deep learning. However, even the most advanced techniques still face challenges in predicting multiple physical properties of conventional composites like plastics, metal alloys, and rubbers, commonly used in everyday life. One example of advanced classical simulations is high-throughput simulations, which accelerate ab initio calculations using efficient algorithms and advanced computational resources to calculate electronic states for billions of atoms and various physical properties of polymers.

In this paper, we view a policy or plan as a transition system over a space of information states that reflect a robot's or other observer's perspective based on limited sensing, memory, computation, and actuation. Regardless of whether policies are obtained by learning algorithms, planning algorithms, or human insight, we want to know the limits of feasibility for given robot hardware and tasks. Toward the quest to find the best policies, we establish in a general setting that minimal information transition systems (ITSs) exist up to reasonable equivalence assumptions, and are unique under some general conditions. We then apply the theory to generate new insights into several problems, including optimal sensor fusion/filtering, solving basic planning tasks, and finding minimal representations for feasible policies.

Paris, May 27, 2021 – Atos and Thales announce the creation of Athea, a joint venture that will develop a sovereign big data and artificial intelligence platform for public and private sector players in the defence, intelligence and internal state security communities. Athea will draw on the experience gained by both companies from the demonstration phase of the ARTEMIS programme, the big data platform of the French Ministry of Armed Forces. The contract to optimise and prepare the full-scale roll-out of the ARTEMIS platform was also awarded jointly to the two leaders by the French Defense Procurement Agency on April 30, 2021. The new joint venture will initially serve the French market before addressing European requirements at a later date. With the exponential rise in the number of sources of information, and increased pressure to respond more quickly to potential issues, State agencies need to manage ever-greater volumes of heterogeneous data and accelerate the development of new AI applications where security and sovereignty are key.

Modern neural network architectures often generalize well despite containing many more parameters than the size of the training dataset. This paper explores the generalization capabilities of neural networks trained via gradient descent. We develop a data-dependent optimization and generalization theory which leverages the low-rank structure of the Jacobian matrix associated with the network. Our results help demystify why training and generalization is easier on clean and structured datasets and harder on noisy and unstructured datasets as well as how the network size affects the evolution of the train and test errors during training. Specifically, we use a control knob to split the Jacobian spectum into "information" and "nuisance" spaces associated with the large and small singular values. We show that over the information space learning is fast and one can quickly train a model with zero training loss that can also generalize well. Over the nuisance space training is slower and early stopping can help with generalization at the expense of some bias. We also show that the overall generalization capability of the network is controlled by how well the label vector is aligned with the information space. A key feature of our results is that even constant width neural nets can provably generalize for sufficiently nice datasets. We conduct various numerical experiments on deep networks that corroborate our theoretical findings and demonstrate that: (i) the Jacobian of typical neural networks exhibit low-rank structure with a few large singular values and many small ones leading to a low-dimensional information space, (ii) over the information space learning is fast and most of the label vector falls on this space, and (iii) label noise falls on the nuisance space and impedes optimization/generalization.

On Friday, November 2nd, Newhouse hosted a panel called Artificial Intelligence & Journalism: Consequences and Opportunities in Emerging Tech. The panelists discussed how artificial intelligence (AI) is moving the technology industry to media organizations. This is seen in the launch of magazine chatbots and daily news organizations use of AI to automate articles. However, it did expose me to research, researchers, and systems that I am excited to explore. In her presentation titled Artificial Unintelligence, Meredith Broussard highlighted the effects of homogeneity in the tech industry, which is mostly white males.

Knowledge workers perform work on many tasks per day and often switch between tasks. When performing work on a task, a knowledge worker must typically search, navigate and dig through file systems, documents and emails, all of which introduce friction into the flow of work. This friction can be reduced, and productivity improved, by capturing and modeling the context of a knowledge worker’s task based on how the knowledge worker interacts with an information space. Captured task contexts can be used to facilitate switching between tasks, to focus a user interface on just the information needed by a task and to recommend potentially other useful information. We report on the use of task contexts and the effect of context on productivity for a particular kind of knowledge worker, software developers. We also report on qualitative findings of the use of task contexts by a more general population of knowledge workers.

Concept Analysis provides a principled approach to effective management of wide area information systems, such as the Nebula File System and Interface. This not only offers evidence to support the assertion that a digital library is a bounded collection of incommensurate information sources in a logical space, but also sheds light on techniques for collaboration through coordinated access to the shared organization of knowledge.

Social media and networking websites, such as Twitter and Facebook, generate large quantities of information and have become mechanisms for real-time content dissipation to users. An important question that arises is: how do we sample such social media information spaces in order to deliver relevant content on a topic to end users? Notice that these large-scale information spaces are inherently diverse, featuring a wide array of attributes such as location, recency, degree of diffusion effects in the network and so on. Naturally, for the end user, different levels of diversity in social media content can significantly impact the information consumption experience: low diversity can provide focused content that may be simpler to understand, while high diversity can increase breadth in the exposure to multiple opinions and perspectives. Hence to address our research question, we turn to diversity as a core concept in our proposed sampling methodology. Here we are motivated by ideas in the "compressive sensing" literature and utilize the notion of sparsity in social media information to represent such large spaces via a small number of basis components. Thereafter we use a greedy iterative clustering technique on this transformed space to construct samples matching a desired level of diversity. Based on Twitter Firehose data, we demonstrate quantitatively that our method is robust, and performs better than other baseline techniques over a variety of trending topics. In a user study, we further show that users find samples generated by our method to be more interesting and subjectively engaging compared to techniques inspired by state-of-the-art systems, with improvements in the range of 15--45%.