We propose the Moral Compass benchmark, a point of reference for incorporating ethical cognition in AI. It has four key contributions. A Moral Decision Dataset (MDD) that captures cases with ethical ambiguity, along with parameters that aid moral decision-making. It is created using a methodology that leverages the use of Large Language Models (LLMs) and seed data from real-world sources which are processed, summarized, and augmented. We also introduce a Moral Decision Knowledge Graph (MDKG) that is created using feature mappings of the relational dataset MDD to facilitate efficient querying. To demonstrate the validity and robustness of this dataset, we introduce an Ethics Scoring Algorithm (ESA) that makes use of the parameters defined in the dataset to calculate ethical scores for isolated actions. Furthermore, ESA is extended by the novel concept of context-sensitive thresholding (CST) to discretize grey areas to resolve ethical dilemmas with explainable results. This work aims to facilitate ethical cognition in AI systems that are deployed in various important sections of society through a clear methodology, modular development, and broad applicability.

Social Synchrony is an important catalyst for learning environments. For a learner, learning outcomes can be greatly enhanced by associating with other compatible peers as part of their learning journey. In many learning environments, learners often autonomously use certain heuristics to connect with other peers. Understanding how these heuristic connections lead to emergent properties at the network level, is important to design learning interventions. In this work, we build simulation models to study the network-level impact of different kinds of heuristics used by learners to form connections. We consider four different heuristics and show their impact both on learning outcomes, and the bookkeeping cost posed by the connections on the learners.

Healthcare accessibility continues to be an important issue in heterogeneous and rapidly growing areas such as Karnataka, India.In these settings, geographical, infrastructural, and socio-economic conditions complicate the access barriers to uniform services.Most conventional accessibility determinations are based on simplistic measures such as Provider-Population ratios or Euclidean distance, which do not suit the multi-faceted aspect of healthcare access.In this research, we introduce GeoHealth Karnataka, a new endto-end geospatial framework that incorporates advanced spatial analysis methods to assess healthcare accessibility holistically.The proposed method integrates high-resolution population rasters, mapping of healthcare facilities, travel-time estimation with friction surfaces, and advanced accessibility metrics such as the Enhanced Two-Step Floating Catchment Area (E2SFCA) approach.The framework surpasses traditional distance-based metrics by adding travel impedance, facility ZoIs (Zone of Influence), and socioeconomic factors in the calculations for accessibility.With an open-source GIS foundation, the implementation highlights marked advancements in detecting gaps across Karnataka's various districts.The system provides visualization and policy-meaningful measures that have the direct ability to inform resource allocation choices.Early case studies identify significant urban-rural differences and identify key areas for intervention, with the potential to revolutionise healthcare planning for more than 70 million residents.

Knowledge graphs are used for organizing and connecting individual entities to integrate the information extracted from different data sources. Typically, knowledge graphs are used to connect various real-world entities like persons, places, things, actions, etc. For the knowledge graphs created using the enterprise data, the knowledge graph entities can be of different types-static entities (e.g., people, projects), communication entities (e.g., emails, meetings, documents), derived entities (e.g., rules, definitions, entities from emails), etc. The graphs are used to connect these entities with enriched context (as edges and node attributes) and used for powering various search and recommendations applications.

There is an increasing demand to design policy interventions to achieve various targets specified by the UN Sustainable Development Goals by 2030. Designing interventions is a complex task given that the system may often respond in unexpected ways to a given intervention. This could be due to interventions towards a given target, affecting other unrelated variables, and/or interventions leading to acute disparities in nearby geographic areas. In order to address such issues, we propose a novel concept called Stress Modeling that analyzes the holistic impact of a policy intervention by taking into account the interactions within a system, after the intervention. The simulation is based on the postulate that complex systems of interacting entities tend to settle down into "low energy'' configurations by minimizing differentials in capabilities of neighbouring entities. The simulation shows how policy impact percolates through geospatial boundaries over time and can be applied at any granularity. The theory and the corresponding package have been explained along with a case study analyzing a fertilizer policy in the Agro-climatic Zones of the state of Karnataka, India.

Education is an inherently social activity. People like to exchange thoughts and learn from each other: this is why we are interested in discussions. But discussions can be messy and vague. In order to make discussions meaningful, relevant mediations may need to be made, whenever discussions lose clarity. Discussion forum data is in the form of a sequence of questions, answers, and comments on the answers. Taken together, this data is called an affirmation. Knowing the clarity of an affirmation makes it possible to intervene in a discussion to steer it towards agreement or conclusion. We have built a model of clarity for a branch of an affirmation which is considered based on scores of agreements, disagreements, and partial answers. We used three classifier models in our experimentation, but the random forest classifier model gave an F1 score of 0.7315, which was better than the other two classifier models. For our model, the ratio of strong agreement to partial agreement was 1.44 for training data and 1.37 for testing data, which is quite close. In the future, we are planning to enhance our model at the sentence level to capture more details and find an aggregate score of clarity for an affirmation.

Detection of online subversive activities, such as fake news, concerted campaigns, and bots, is getting increasingly urgent. However, without specific knowledge of underlying facts and disparate valid perspectives of a given issue, it is hard to detect subversive intent in a generic sense. To address this, we approach the problem from a “macro” perspective. Rather than asking whether a specific social media account is acting subversively, we look at the entire discourse around a trending topic, and ask whether the discourse looks “healthy” or is it showing signs of getting hijacked or dominated by one particular perspective. To do this, we break down a social discourse into its constituent narratives. Narratives are in turn modeled as latent stories or worldviews, whose visible characterizations are in the form of specific distributions over different opinions expressed in the discourse. Once the discourse is broken down into narratives, the “health” of the discourse can be addressed using various measures, such as the relative sizes of its constituent narratives, sentiment polarity of internarrative interactions, and presence or absence of dominant players within each narrative. We conduct experiments on several well-known trending topics on Twitter to identify its constituent narratives and provide a report card on the overall discourse quality. We also show how this top-down approach offers the means to delineate roles played by users as drivers of the discourse, the constituent narratives, or their component opinions, determined on the basis of dominance centrality measures and narrative affinities.

The web is seen as a promising platform for designing scalable educational practices across large populations. Many of the efforts in this space use the web primarily as an amplifier over existing models of learning that are based on the classroom. In this paper, we propose a pedagogic model called mediated learning where the web acts as a platform for nurturing a learning community by continuously mediating between knowledge need and expertise. Mediated learning has the potential to invert the learning pyramid by interfacing the learner with several experts as part of a single learning experience. For supporting mediated learning, an approach is needed that is data-intensive and driven by social semantics. This paper outlines the proposed pedagogic model, which comprises two primary components: a user-end navigator component that provides a rich interface enabling users to independently navigate through a learning space; and a back-end community component, that performs meaningful mediations between participants in the logical learning space.

There are a number of inbound web services, which recommend content to users. However, there is no way for such services to prioritize their recommendations as per the users' interests. Here we are not interested in generating new recommendations, but rather organizing and prioritizing existing recommendations in order to increase the click rate. Since users have different patterns of browsing that also change frequently, it is good to have a system that prioritizes recommendations based on the current browsing patterns of individual users. In this paper we present such a system. We first generate the clusters of article topics using URLs from the users' browsing history, which is then used to generate the relevancy scores of the recommendation services based on entropy. The relevancy scores are then fed to the service providers, which use them to prioritize their recommendations by ranking them based on the relevancy scores. We test the model using the browsing history for 10 users, and validate the model by calculating the correlation of the generated relevancy scores with the users' manually provided topic preferences. We further use collaborative filtering to benchmark the usefulness of our ranking systems.

Lexical cooccurrence in textual data is not uniformly random. The statistics inferred from the term-cooccurrence data enable us to model dependencies between terms as graphs, somewhat resembling the way semantic memory is organised in human beings. In this paper we look at cooccurrence patterns to identify topical anchors of a given context. Topical anchors are those terms whose semantics represent the topic of the whole context. This work is based on computing a stationary distribution in the cooccurrence graph. Topical anchors were computed on a set of 100 contexts and were also evaluated by 86 volunteers and the results show that the algorithm correctly identifies the topical anchors around 62% of the time.

Stream grids are wide-area grid computing environments that are fed by a set of stream data sources. Queries arrive at the grid from users and applications external to the system. The kind of queries considered in this work are long-running continuous (LRC) queries, that we also term as "open-world" queries. These queries are neither short-lived nor infinitely long lived. They live long enough to make the prospect of multi-query optimization meaningful. But queries may also terminate at any time, requiring re-optimization of the query plans. The queries are "open" from the grid perspective as the grid cannot control or predict: (1) arrival of a query with time, location, required data and, (2) query revocations. Query optimization in such an environment has two major challenges: (a) optimizing in a multi-query environment and (b) continuous optimization due to new query arrivals and revocations. As generating a globally optimal query plan is an intractable problem, this work explores the idea of emergent optimization, where globally optimal query plans emerge as a result of local autonomous decisions taken by the grid nodes. Drawing concepts from evolutionary game theory, grid nodes are modeled as autonomous agents that seek to maximize a self-interest function using one of a set of different strategies. Grid nodes change strategies in response to variations in query arrival and revocation patterns. Changing of strategies is also autonomously decided by each grid node based on how its strategy is faring with respect to other strategies in the grid.

We address the problem of designing optimal network topologies under arbitrary optimality requirements. Using three critical system parameters, efficiency, robustness and cost, we evolve optimal topologies under different environmental conditions. Two prominent classes of topologies emerge as optimal: (1) Star-like topologies, with high efficiency, high resilience to random failures and low cost, and (2) ¿Circular Skip Lists¿ (CSL), with high robustness to random failures as well as targeted attacks, and high efficiency at moderate cost. We analyze CSLs to observe that they show several structural motifs that are optimal with respect to a variety of metrics.

Query optimization in sensor grids have two major challenges: (a) optimizing in a multi-query environment, and (b) continuous re-optimization occurring due to new query registrations and de-queries, i.e. queries being stopped unexpectedly. Addressing this problem continuously on a system-wide basis is an infeasible option. In this work called EstuaryDB, we propose a notion of emergent optimization, where globally optimal configurations emerge as a result of a number of local autonomous decisions carried out in self-interest. Grid nodes act as self-interested autonomous agents that continuously seek to maximize their "wealth." The agents are unaware of system-wide issues such as when do queries arrive, what are they asking for, or when are they revoked. Every query brings with it a certain amount of wealth, and each agent continuously tries to save as much of the wealth as possible. The amount of latent wealth in the system at any time gives a quantitative measure of the efficiency achieved over naive stream retrieval.

Large information systems (IS) comprise of several independent applications that share a common set of resources and data. Usually, there are implicit and subtle dependencies across these applications that are not specifically captured. This is especially so if the applications are bought off the shelf or are developed by independent third parties. Dependencies or global semantic constraints are difficult to discern and incorporate into the design of individual software components. Global constraints may change over time and it is usually expensive or infeasible to change individual application logic in every such situation. In order to address such an issue, we propose LogicFence, a framework that accepts a definition of global constraints and translates these constraints into primitives that are embedded into the run-time environments of application programs (currently, into the JVM of Java applications). LogicFence monitors the state of application programs and prevents the disparate instances to collectively form a globally inconsistent state

This demonstration presents a database system called GRACEfor storageand retrieval of graphstructures. Structural queries are supported which retrieve graphs based on approximate subgraph isomorphism. Since subgraph isomorphism is NP-complete, GRACE performs retrieval based on inexact graph matching. The underlying model is of a concept called “Vectorization of Structure” that represents structural features of member graphs as vectors in one or more hypothetical spaces. Queries are mapped onto regions in these spaces. Query results is a ranked union of the set of all points lying in the query regions. The implementation displays a GRACE model implemented for storage and retrieval of molecular structures of organic chemicals.

Providers of Web based services are interested in monitoring the usage of their services in combination with those of other providers. The identification of services frequently accessed together may be valuable as a basis for strategic collaboration among their owners. We propose data mining to discover services of different providers which could complement one another, based on their usage. In particular we model the activities of a user as a sequence of service invocations recorded in a log, on which pattern discovery techniques can be applied. However we claim that conventional sequence mining is not adequate for this type of application. This is because, conventional mining concentrates on frequent (or infrequent) patterns of access, while we also require a notion of the consistency of these access patterns as a basis for collaboration. We present a model for constructing patterns that depict consistently used sequences of activities. This model is general enough to be applied to any system of autonomous entities, where relationships between entities are dynamic. For testing our model, we have analyzed the behavior of users in a news group, in order to determine consistent patterns in the way users respond to questions posed to the group.