arXiv subjects

Karthik Shyamsunder

Publications and source records attributed to Karthik Shyamsunder.

3 recordsLinked to original sources

arXiv2026

DOGS: Design-Space Sampling for Prompt-Driven Logo Generation

Prompt optimization for text-to-image (T2I) generation has been pursued almost entirely as text rewriting, in which a short user brief is expanded into a longer, model-preferred token sequence. We argue that such a language-space formulation is ill-suited to structured visual design tasks such as logo creation, where a one-line brief leaves most design decisions unspecified. These decisions depend on relational priors that a linear sequence cannot encode, and they leave an uncontrolled channel through which protected marks may be reproduced. We therefore recast logo prompting as sampling within a structured design space, and instantiate this idea as DOGS (Design-space prompting with an Originality-aware GFlowNet Sampler). From a large corpus of real-world logos, we mine a typed, graph-structured design grammar whose edges record empirical co-occurrence. A GFlowNet sampler then generates design graphs with probability proportional to a terminal reward that combines recognizability, aesthetics, and corpus-relative originality. Every slot draws only from a closed design-level vocabulary, and any infringement-inducing or harmful token is removed during parsing. The originality reward further penalizes proximity to existing logos, thereby incorporating infringement avoidance into the method by construction. On two open-source renderers and against nine baselines, DOGS produces logos that are more recognizable and aesthetic, substantially more diverse, and far less prone to trademark infringement.

cs.CV↗

arXiv2026

Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery

Removing copyrighted, unsafe, or user-specified concepts from a deployed text-to-image diffusion model is now a practical requirement. Weight-editing methods can suppress fixed targets, but they require per-target retraining and modify the model checkpoint. Training-free methods, on the other hand, are deployment-friendly, but they suffer from text-side routing failures on compositional prompts. In such prompts, the erase target may be invoked through a related class rather than its lexical name, and its modifiers may migrate onto preserved objects. This paper proposes Hierarchically Grounded Semantic Surgery (HGSS), a training-free framework for compositional concept erasure. The framework lifts both the routing signal and the edit operator used by text-side erasure. First, hierarchical span grounding resolves erase-target spans through lexical, taxonomic, and semantic evidence, while guarding against broad-hypernym and compound-head false positives. Second, dynamic attribute binding refines the text conditioning during early denoising via a counterfactual reference and a preserve-aware cross-attention objective, keeping surviving attribute-noun bindings intact. HGSS selectively removes the erase target without updating model weights or adding learned parameters. On SEE, HGSS cuts hierarchical evasion from 29.54 to 10.02 and roughly halves pairwise attribute leakage, achieving the best Neighbor E and AttrP scores among the reported erasure methods. On UnlearnCanvas, HGSS slightly improves the six-metric average over the matched Semantic Surgery baseline, reaching state-of-the-art.

cs.CV↗

arXiv2026

Efficient and Secure Discovery for AI Agents: The Case for DNS

As Artificial Intelligence (AI) agents enter their next stage of ubiquitous deployment throughout the Internet, their discoverability will become a central challenge. The information AI agents need to discover one another, how they will locate it, how to facilitate authentication, integrity, and authorization, how to connect across different platforms, and how to scale across organizational boundaries form a set of unanswered challenges that will be central to deployment success. These are challenges for which one of the Internet's most venerable, solid, and ubiquitous infrastructures is ideally suited: The Domain Name System (DNS). Such a rich, already ubiquitous, and programmatically flexible foundation is an ideal option for discovery of AI agents. In this work, we propose an illustration and rationale for the basic semantics that discovery for AI agents will require. We argue that three key evaluation criteria will become paramount: navigational completeness (the extent to which the necessary metadata, with elements like trust, is included in a discovery solution), lookup complexity, and transaction performance (e.g., latency, speed, or recency). Using 119,757 real-world service endpoints, 62,739 indexed MCP server entries, and multiple agent tooling ecosystems, we empirically evaluate the first of these considerations to illustrate the appropriateness of using DNS for AI agent discovery. We find that the necessary and sufficient discovery metadata can be encoded into a single DNS resource record that fits within one DNS UDP response, enabling single-RTT discovery. Our evaluations illustrate a promising path toward enabling AI agent discoverability at the Internet's scale, and thereby accelerating secure, stable, and resilient deployments.

cs.NI↗