Research

Examines how language models narrow the set of career options they present, and what that narrowing does to the people relying on them for guidance.

Measures how streaming fragmentation relates to earlier leaks, larger torrent swarms, and increased risk to the users who join them.

Studies how online word-of-mouth on Glassdoor shapes organizational image and applicant behavior in recruitment contexts.

Measures attacks on node infrastructure, models how failures cascade through the network, and evaluates tactics for rapid recovery.

Compares human and agentic prompt refinement on the same image-regeneration task, and locates where the two diverge.

A perceptual similarity metric tuned against human judgments, so similarity scoring is usable as feedback inside iterative generation workflows.

Examines how AI-generated employer summaries shape employer image, candidate confidence, and the decision to apply.

Evaluates human ability to infer the prompts used to generate AI art. Findings show that prompt inference is difficult and that contextual factors significantly affect performance.

Presents NinjaDoH, a moving-target DNS-over-HTTPS architecture that combines hyperscaler deployment with IPNS-based endpoint agility to improve censorship resistance and availability.

Proposes a taxonomy for evaluating dataset value in AI-native empirical research. Moves beyond volume-centric data paradigms to consider dimensions of quality, representativeness, and ethical sourcing.

Evaluates the efficacy of iterative human-driven prompt refinement in AI image regeneration tasks. Findings reveal both the potential and limits of human ability to reverse-engineer prompts from generated images.

Documents continuous, automated scanning and fingerprinting of Ethereum nodes, with geographic and temporal structure in the reconnaissance activity.

A fifteen-year torrent-discovery dataset of 4,936,257 magnet links matched to IMDb metadata, with daily snapshots of more than 130,000 unique torrents.

Couples global node performance measurement with honeypot interactions and full network sessions, so blockchain infrastructure security can be studied reproducibly.

Audits the privacy and data-sharing practices behind email marketing at large companies, and what those third-party flows mean for the people receiving the mail.

The measurement dataset behind the CODASPY study of email marketing and inbox privacy.

Public release of the torrent-discovery dataset with IMDb-matched movies and TV shows.

Realtime surveillance and changepoint-based anomaly detection for Ethereum P2P security.

Purpose-built VPN overlay to obscure node IPs and improve resilience against network-layer attacks.

Decentralized storage pipeline for faster Ethereum snapshot delivery and recovery.

Node telemetry, full network sessions, honeypot interactions, and logs from the EtherBee measurement infrastructure.

Proposes NFT-anchored, trust-minimized validator identity and onboarding for proof-of-stake networks.

Applies NLP and bibliometric analysis to the pandemic literature, mapping research clusters, collaboration networks, and how topics moved over time.

Tests whether employer branding and attitudes toward government predict which students apply to public-sector technical roles.

Book review for Air & Space Power Journal.

Book review for Air & Space Power Journal.

HR managers have a critical but underrecognized role in defending organizations against social engineering. This article argues that HR's people-facing position makes them essential for awareness, training, and policy.