Player decision-making in multiplayer card games presents opportunities to improve human interaction with complex systems. We introduce GNNetic , a Relational Graph Convolutional Network for game analysis in card games. By encoding card relationships and game context as a graph, GNNetic frames each discard choice in Rummy (case study) as a graph-classification task learned from expert play, enabling the identification of players’ action deviations. We leverage these insights to address core HCI challenges in game design, interface optimization, and player development. GNNetic helps surface recurring patterns of suboptimal play that may indicate design-induced errors, informing iterative refinements to enhance usability and reduce cognitive load during gameplay. Our analysis and summative study demonstrate this potential: UI enhancements informed by GNNetic ( Companion Mode in practice games) reduce common errors like joker discards by ~31% and invalid declarations by ~27% . We also propose Skill Arena —a post-game visualization that renders these deviations accessible, providing feedback to foster metacognitive awareness and facilitate skill enhancement. GNNetic achieves a ~15x reduction in model size with ~6.5% accuracy improvement over larger baselines. By bridging expert strategy with interface design, this work paves the way for more intuitive UIs and targeted skill enhancement in card games, demonstrating tangible performance gains through interface improvements.

Can we automatically predict failures of an object detection model on images from a target domain? We characterize errors of a state-of-the-art object detection model on the currently popular smart mobility domain, and find that a large number of errors can be identified using spatial commonsense. We propose øurmodel , a system that automatically identifies a large number of such errors based on commonsense knowledge. Our system does not require any new annotations and can still find object detection errors with high accuracy (more than 80% when measured by humans). This work lays the foundation to answer exciting research questions on domain adaptation including the ability to automatically create adversarial datasets for target domain.

Leveraging social networks in computer systems can be effective in dealing with a number of trust and security issues. Spam is one such issue where the "wisdom of crowds" can be harnessed by mining the collective knowledge of ordinary individuals. In this paper, we present a mechanism through which members of a virtual community can exchange information to combat spam. Previous attempts at collaborative spam filtering have concentrated on digest-based indexing techniques to share digests or fingerprints of emails that are known to be spam. We take a different approach and allow users to share their spam filters instead, thus dramatically reducing the amount of traffic generated in the network. The resultant diversity in the filters and cooperation in a community allows it to respond to spam in an autonomic fashion. As a test case for exchanging filters we use the popular SpamAssassin spam filtering software and show that exchanging spam filters provides an alternative method to improve spam filtering performance.

In order for autonomic systems to function, the individual components must co-operate and not indulge in malicious behavior. However, it is almost certain that autonomous systems in. Next generation networks will inadvertently include less than trustworthy components. Identifying such entities is critical to the smooth and effective functioning. We present new experiments conducted with the ROCQ scheme, a reputation-based trust management system that computes the trustworthiness of peers on the basis of transaction-based feedback. The ROCQ model combines four parameters: reputation (R) or a peer's global trust rating, opinion (O) formed by a peer's first-hand interactions, credibility (C) of a reporting peer and quality (Q) or the confidence a reporting peer puts on the feedback it provides. In this paper, we demonstrate that ROCQ is robust against churn and also examine the effect of credibility and quality on the performance of the scheme.

In this paper, we propose a swarm intelligence strategy for constructing a mobile backbone multicasting network which leads to improved scalability and connectivity compared to conventional flat networks. The strategy combines a simple clustering technique and on demand multicast protocol (ODMRP). Also we show benefits of the mobile backbone network through a simulation study.

We propose a reputation-based trust management system, ROCQ, to reduce unauthentic and corrupted file transfers in end-user collaborative content-distribution systems. Such systems are characterized by the splitting of large files into (optionally encoded) blocks and the simultaneous downloading of several blocks from different nodes to speed up content distribution. All nodes must cooperate and provide correct content for such systems to function smoothly. If malicious nodes are present, they can quickly bring the system to a halt by introducing fake blocks in the network making it impossible to reconstruct the original file. ROCQ (reputation, opinion, credibility and quality) uses feedback from past interactions among nodes to create node reputations. This allows detection of malicious nodes in a transaction-based network. We study the performance of ROCQ as used in a content distribution network. We find that ROCQ decreases the likelihood of a user receiving fake blocks by up to 16%, and, hence, significantly reduces bandwidth waste.