Carnegie Mellon University

Technology can make the world a better place. But it must be done so carefully and with a fuller understanding of the implications it brings.

What should the infrastructure for a smart, connected city of the future look like? Does it adequately address the real problem of sustainability? What functionality does that infrastructure expose and who controls it?

Whether it building smart grids which use historical usage data to estimate consumption or promoting public health through the medicalization of off-the-shelf consumer electronics, an understanding of the problem being addressed and the challenges the solution gives rise to is critical.

It is here that our work on applied systems and infrastructure thrives. In a world where sensors are embedded in our roads and Internet of Things technologies generate massive amounts of data, how do we engineer such systems to do the most good while preventing misuse? Our cross-cutting faculty address such concerns in an interdisciplinary fashion, leveraging computer science, big data analysis, electrical and computer engineering, alongside social sciences to forge breakthroughs which will shape the world of responsibly developed technology to come.

In a world where sensors are embedded in our roads and Internet of Things technologies generate massive amounts of data, how do we engineer such systems to do the most good while preventing misuse?

Example Research

Training Robust ML-based Raw-Binary Malware Detectors in Hours, not Months

This groundbreaking work tackles a critical challenge in cybersecurity: how to make machine learning-based malware detection both robust against evasion attacks and practically deployable at scale. While previous approaches required months of intensive computation to train effective malware detectors, the researchers developed innovative techniques that achieve superior results in mere hours, democratizing access to robust malware detection capabilities. The work exemplifies how clever algorithmic design, informed by deep understanding of both technical and practical constraints, can transform seemingly intractable problems into manageable ones. This research demonstrates the kind of impactful work done in Societal Computing that combines rigorous technical innovation with real-world applicability to make sophisticated security tools more accessible to organizations of all sizes.

Learn more

Function Approximation for Solving Stackelberg Equilibrium in Large Perfect Information Games

Function approximation (FA) has been pivotal in large zero-sum games, but limited work explores its use in general-sum extensive-form games, which are computationally challenging. This paper introduces the Enforceable Payoff Frontier (EPF), a generalization of the state value function for general-sum games, approximated via neural networks to solve for the Stackelberg extensive-form correlated equilibrium (SEFCE) in two-player games of perfect information. This approach scales to larger games, provides incentive compatibility guarantees, and achieves high performance without relying on self-play or approximate best-response oracles. Experimental results demonstrate the method's effectiveness in approximating solutions for complex games beyond traditional traversal methods.

Learn more

CodeWalk: Facilitating Shared Awareness in Mixed-Ability Collaborative Software Development

For blind and visually impaired (BVI) developers, participating in real-time code collaboration has long been challenging due to accessibility barriers. This paper introduces CodeWalk, a tool that integrates with Visual Studio Code's Live Share extension to make remote, synchronous coding sessions more accessible. By using sound effects and speech cues to convey navigation and editing actions, CodeWalk empowers BVI developers to stay engaged and reduce the need for frequent updates from collaborators. Testing shows that CodeWalk significantly eases coordination, fostering more active participation for BVI developers in collaborative coding.

Learn more

PrISM-Observer: Intervention Agent to Help Users Perform Everyday Procedures Sensed using a Smartwatch

We routinely perform procedures (such as cooking) that include a set of atomic steps. Often, inadvertent omission or misordering of a single step can lead to serious consequences, especially for those experiencing cognitive challenges such as dementia. This paper introduces PrISM-Observer, a smartwatch-based, context-aware, real-time intervention system designed to support daily tasks by preventing errors...

Learn more

NewsPanda: Media Monitoring for Timely Conservation Action

NewsPanda is an automated toolkit designed to monitor conservation and infrastructure media for NGOs. Using a fine-tuned BERT model, it identifies and analyzes relevant articles, providing timely updates. Deployed by WWF, it scans thousands of sites globally, significantly reducing manual monitoring efforts...

Learn more

MITES: A Privacy-Aware General-Purpose Sensing Infrastructure for Smart Buildings

The Mites project tackles one of the most complex challenges in smart building technology: how to deploy comprehensive sensing infrastructure while respecting privacy, ensuring security, and maintaining community trust in a shared space. Through an innovative combination of hardware design, privacy-preserving architecture, and extensive community engagement, the researchers developed a system that successfully balances the competing needs of different stakeholders - from building managers seeking efficiency to occupants concerned about surveillance. This work exemplifies how technical innovation must be deeply integrated with social considerations, as demonstrated by their iterative design process that incorporated community feedback and led to novel solutions like location obfuscation to protect occupant privacy. The project showcases the kind of interdisciplinary thinking central to Societal Computing, where cutting-edge technical solutions are shaped by and responsive to human needs and social dynamics...

Learn more

TAO: Context Detection from Daily Activity Patterns Using Temporal Analysis and Ontology

TAO is a hybrid system that translates detailed activity detection into meaningful contexts for healthcare and human-computer interaction applications. Using OWL-based ontologies and temporal clustering, it captures complex activity patterns with high accuracy. Real-world deployment demonstrates TAO’s effectiveness for wellness applications, outperforming traditional approaches...

Learn more

The Personalized Privacy Assistant Project

Privacy management is becoming overwhelming as IoT and Big Data expand the ways our information can be collected and used. Personalized privacy assistants offer a solution: AI agents that learn our preferences and help manage privacy settings on our behalf. These assistants would selectively alert users about concerning practices, confirm uncertain decisions, and refine their understanding over time - ultimately helping users maintain control over their data without constant manual intervention...

Learn more