The competitive MNIST handwritten digit recognition benchmark has a long history of broken records since 1998. The most recent substantial improvement by others dates back 7 years (error rate 0.4%) . Recently we were able to significantly improve this result, using graphics cards to greatly speed up training of simple but deep MLPs, which achieved 0.35%, outperforming all the previous more complex methods. Here we report another substantial improvement: 0.31% obtained using a committee of MLPs.

We define a logic of propositional formula schemata adding to the syntax of propositional logic indexed propositions and iterated connectives ranging over intervals parameterized by arithmetic variables. The satisfiability problem is shown to be undecidable for this new logic, but we introduce a very general class of schemata, called bound-linear, for which this problem becomes decidable. This result is obtained by reduction to a particular class of schemata called regular, for which we provide a sound and complete terminating proof procedure. This schemata calculus allows one to capture proof patterns corresponding to a large class of problems specified in propositional logic. We also show that the satisfiability problem becomes again undecidable for slight extensions of this class, thus demonstrating that bound-linear schemata represent a good compromise between expressivity and decidability.

In large systems, it is important for agents to learn to act effectively, but sophisticated multi-agent learning algorithms generally do not scale. An alternative approach is to find restricted classes of games where simple, efficient algorithms converge. It is shown that stage learning efficiently converges to Nash equilibria in large anonymous games if best-reply dynamics converge. Two features are identified that improve convergence. First, rather than making learning more difficult, more agents are actually beneficial in many settings. Second, providing agents with statistical information about the behavior of others can significantly reduce the number of observations needed.

The development of discursive knowledge presumes the communication of meaning as analytically different from the communication of information. Knowledge can then be considered as a meaning which makes a difference. Whereas the communication of information is studied in the information sciences and scientometrics, the communication of meaning has been central to Luhmann's attempts to make the theory of autopoiesis relevant for sociology. Analytical techniques such as semantic maps and the simulation of anticipatory systems enable us to operationalize the distinctions which Luhmann proposed as relevant to the elaboration of Husserl's "horizons of meaning" in empirical research: interactions among communications, the organization of meaning in instantiations, and the self-organization of interhuman communication in terms of symbolically generalized media such as truth, love, and power. Horizons of meaning, however, remain uncertain orders of expectations, and one should caution against reification from the meta-biological perspective of systems theory.

Organizational agility is a key challenge in today's business world. The Knowledge-Intensive Service Support approach tackles agility by combining process modeling and business rules. In the paper at hand, we present five approaches of process mining that could further increase the agility of processes by improving an existing process model.

While much health data is available online, patients who are not technically astute may be unable to access it because they may not know the relevant resources, they may be reluctant to confront an unfamiliar interface, and they may not know how to compose an answer from information provided by multiple heterogeneous resources. We describe ongoing research in using natural English text queries and automated deduction to obtain answers based on multiple structured data sources in a specific subject domain. Each English query is transformed using natural language technology into an unambiguous logical form; this is submitted to a theorem prover that operates over an axiomatic theory of the subject domain. Symbols in the theory are linked to relations in external databases known to the system. An answer is obtained from the proof, along with an English language explanation of how the answer was obtained. Answers need not be present explicitly in any of the databases, but rather may be deduced or computed from the information they provide. Although English is highly ambiguous, the natural language technology is informed by subject domain knowledge, so that readings of the query that are syntactically plausible but semantically impossible are discarded. When a question is still ambiguous, the system can interrogate the patient to determine what meaning was intended. Additional queries can clarify earlier ones or ask questions referring to previously computed answers. We describe a prototype system, Quadri, which answers questions about HIV treatment using the Stanford HIV Drug Resistance Database and other resources. Natural language processing is provided by PARC’s Bridge, and the deductive mechanism is SRI’s SNARK theorem prover. We discuss some of the problems that must be faced to make this approach work, and some of our solutions.

GENE_ARCH is a Generative Design System that combines Pareto Genetic Algorithms with an advanced energy simulation engine. This work explores its integration with a Shape Grammar, acting as GENE_ARCH’s shape generation module. The islamic patio house typology is readdressed in a contemporary context, by improving its energy-efficiency, and rethinking its role in the genesis of high-density urban areas, while respecting its specific spatial organization and cultural grounding. Field work was carried out in Marrakesh, surveying a number of patio houses, becoming the Corpus of Design, from where a shape grammar was generated. The computational implementation of the patio house grammar was done within GENE_ARCH. The resulting program was able to generate new, alternative patio houses designs that were more energy efficient, while respecting the traditional rules captured from the analysis of existing houses. After the computational system was fully implemented, it was possible to realise a large number of experiments. The first experiments kept more restrained rules, thus generating new designs that closer resembled the existing ones. The progressive relaxation of rules and constraints allowed for a larger number of variations to emerge. Analysis of energy results provide insight into the main patterns resulting from the GA search processes.

While computers assist humans with tasks such as navigation that involve spatial aspects, agents that can interact in a meaningful way in this context are still in their infancy. One core issue is the mismatch in the representation of spatial information a computer-based system is likely to use, and the one a human is likely to use. Computers are better suited for quantitative schemes such as maps or diagrams that rely on measurable distances between entities. Humans frequently use higher-level, domain-specific conceptual representations such as buildings, rooms, or streets for orientation purposes. Combined with the person-centric world view that we often assume when we refer to spatial information, it is challenging for agents to convert statements using spatial references into assertions that match their own internal representation. In this paper, we discuss an approach that uses natural language processing and information extraction tool kits to identify entities and statements about their spatial relations. These extractions are then processed by a spatial reasoner to convert them from the human conceptual space into the quantitative space used by the computer-based agent.

The collaboration between a human and a robot is here understood as a learning process mediated by the instructor prompt behaviours and the apprentice collecting information from them to learn a plan. The instructor wears the Gaze Machine, a wearable device gathering and conveying visual and audio input from the instructor while executing a task. The robot, on the other hand, is eager to learn both the best sequence of actions, their timing and how they interlace. The cross relation among actions is specified both in terms of time intervals for their execution, and in terms of location in space to cope with the instruction interaction with people and objects in the scene. We outline this process: how to transform the rich information delivered by the Gaze Machine into a plan. Specifically, how to obtain a map of the instructor positions and his gaze position, via visual slam and gaze fixations; further, how to obtain an action map from the running commentaries and the topological maps and, finally, how to obtain a temporal net of the relevant actions that have been extracted. The learned structure is then managed by the flexible time paradigm of flexible planning in the Situation Calculus for execution monitoring and plan generation.

Demographic predictions of population aged 65 and over suggest the need for telemedicine applications in the eldercare domain. Current solutions are mainly focused on fall detection. This paper presents a working prototype of a system that in addition to fall detection monitors a variety of user’s behavior characteristics that help raise awareness of health risks. It monitors the state of user’s health, and more importantly, detects changes in behavior characteristics that potentially indicate a forthcoming or current disease, illness or some other disability. The system utilizes domain knowledge from medical literature on quantitative behavior analysis and combines it with an outlier-detection algorithm in order to identify anomalous behavior. Preliminary results with the working prototype are promising, showing a potential to deploy the system in practice.