CISL is working towards physical, perceptual safety and alignment.

We are a team of researchers currently exploring diffusion architectures, raster-to-vector image conversion, and prototyping to develop and improve generative models for safety applications.

We are an open, collaborative research environment where experimentation and reproducibility guide every stage, from model design and benchmarking to press-ready vector output.

---

Research Interests

• Physically-constrained generative pipelines — diffusion architectures with embedded manufacturing constraints for industrial print and packaging workflows
• Synthetic media forensics — adversarial robustness metrics, cross-domain generalisation, and detection-evasion trade-offs for deepfake identification
• Physics-informed generation — constraint-aware synthesis for CAD/AEC applications including structural topology and fabrication-aware geometry
• Prompt-output alignment — compositional semantics, attribute binding accuracy, and spatial relationship preservation in text-to-image systems
• Privacy-preserving vision — federated learning protocols, differential privacy, and latency-bounded inference for edge deployment