Despite their immense success as a model of macaque visual cortex, deep convolutional neural networks (CNNs) have struggled to predict activity in visual cortex of the mouse, which is thought to be strongly dependent on the animal's behavioral state. Furthermore, most computational models focus on predicting neural responses to static images presented under head fixation, which are dramatically different from the dynamic, continuous visual stimuli that arise during movement in the real world. Consequently, it is still unknown how natural visual input and different behavioral variables may integrate over time to generate responses in primary visual cortex (V1). To address this, we introduce a multimodal recurrent neural network that integrates gaze-contingent visual input with behavioral and temporal dynamics to explain V1 activity in freely moving mice. We show that the model achieves state-of-the-art predictions of V1 activity during free exploration and demonstrate the importance of each component in an extensive ablation study. Analyzing our model using maximally activating stimuli and saliency maps, we reveal new insights into cortical function, including the prevalence of mixed selectivity for behavioral variables in mouse V1. In summary, our model offers a comprehensive deep-learning framework for exploring the computational principles underlying V1 neurons in freely-moving animals engaged in natural behavior.

A semantics-preserving exchange of information between two software applications requires mappings between logically equivalent concepts in the ontology of each application. The challenge of semantic integration is therefore equivalent to the problem of generating such mappings, determining that they are correct, and providing a vehicle for executing the mappings, thus translating terms from one ontology into another. This article presents an approach toward this goal using techniques that exploit the model-theoretic structures underlying ontologies. With these as inputs, semiautomated and automated components may be used to create mappings between ontologies and perform translations. A major barrier to such interoperability is semantic heterogeneity: different applications, databases, and agents may ascribe disparate meanings to the same terms or use distinct terms to convey the same meaning.

This article describes one person's experience in coming from an academic environment to work at Digital Equipment Corpo I've divided this history into two distinct parts. AI and DEC's entry into the AI market, DEC engineers were This article is an edited version of Dr Polit's presentation at the Technology Transfer Symposium held at the AAAI-83 conference Building Expert Systems I'll now give a brief review of the steps involved in building expert systems as they are described by many researchers. The five steps involved in building an expert system are: Step 1: problem recognition, Step 2: task definition, Step 3: initial design, Step 4: knowledge acquisition, and Step 5: system maintenance. Frequently, the problem is perceived as a bottleneck in a larger process; sometimes it is a scarcity of traiued personnel. Second, during step 2, researchers must define the functions the AI system will perform.

People often act together with a shared purpose; they collaborate. Collaboration enables them to work more efficiently and to complete activities they could not accomplish individually. An increasing number of computer applications also require collaboration among various systems and people. Thus, a major challenge for AI researchers is to determine how to construct computer systems that are able to act effectively as partners in collaborative activity. Collaborative activity entails participants forming commitments to achieve the goals of the group activity and requires group decision making and group planning procedures.

The notion of generic design, although it has been around for 25 years, is not often articulated; such is especially true within Newell and Simon's (1972) informationprocessing theory (IPT) framework. Design is merely lumped in with other forms of problem-solving activity. Intuitively, one feels there should be a level of description of the phenomenon that refines this broad classification by further distinguishing between design and nondesign problem solving. However, IPT does not facilitate such problem classification. This article makes a preliminary attempt to differentiate design problem solving from nondesign problem solving by identifying major invariants in the design problem space.

However, interoperability among these manufacturing applications is hindered because the applications use different terminology and representations of the domain. These problems arise most acutely for systems that must manage the heterogeneity inherent in various domains and integrate models of different domains into coherent frameworks (figure 1). For example, such integration occurs in businessprocess reengineering, where enterprise models integrate processes, organizations, goals, and customers. Even when applications use the same terminology, they often associate different semantics with the terms. This clash over the meaning of the terms prevents the seamless exchange of information among the applications.

This article discusses a workshop held in conjunction with the Eighteenth International Joint Conference on Artificial Intelligence (IJCAI-03), held in Acapulco, Mexico, on 11 August 2003. However, much of the work does not take into account real-time constraints typically associated with many agent applications in addition to the incomplete and dynamic nature of the embedding environments. For example, in environments where a number of agents build teams, and both singleagent and collaborative decisions have to be made, such decisions have to be generated rapidly and in the appropriate time windows to be useful. Such topics include world modeling, planning, learning, agent communication, and software architectures. Within this general theme, the aim was to bring together researchers from different communities working with both robots and softbots (for example, RoboCup, cognitive robotics, intelligent autonomous vehicles).

This ease has spurred an increasing interest from professionals, the general public, and consequently politicians to make publicly available the tremendous wealth of information kept in museums, archives, and libraries--the so-called memory organizations. Quite naturally, their development has focused on presentation, such as web sites and interfaces to their local databases. Now with more and more information becoming available, there is an increasing demand for targeted global search, comparative studies, data transfer, and data migration between heterogeneous sources of cultural contents. The reality of semantic interoperability is getting frustrating. In the cultural area alone, dozens of standard and hundreds of proprietary metadata and data structures exist as well as hundreds of terminology systems.

Until quite recently, it was taken for granted in AIand cognitive science more broadlythat activity resulted from the creation and execution of plans. In 1985, several researchers, including myself, independently realized that plans and planning are not necessary-or necessarily useful-in activity. Since this time, a number of alternatives have been proposed. This analysis is equally applicable to any other computational problem. Thus, you could conclude that vision is impossible because it requires exponential computation in the number of pixels or that, on the average, business data processing takes exponential work in the number of records.

We propose adding to the mix a novel architecture, the process-integrated mechanism (PIM), that enjoys the advantages of having a single controlling authority while avoiding the structural difficulties that have traditionally led to the rejection of centralized approaches in many complex settings. In many situations, PIMs improve on previous models with regard to coordination, security, ease of software development, robustness, and communication overhead. In the PIM architecture, the components are conceived as parts of a single mechanism, even when they are physically separated and operate asynchronously. The PIM model offers promise as an effective infrastructure for handling tasks that require a high degree of time-sensitive coordination between the components, as well as a clean mechanism for coordinating the high-level goals of loosely coupled systems. The PIM model enables coordination without the fragility and high communication overhead of centralized control, but also without the uncertainty associated with the system-level behavior of a multiagent system (MAS).