Technical practice
Edge AI Model Optimization
What to coverDiscuss techniques for deploying deep learning models on resource-constrained edge devices. Candidates should cover model compression methods like quantization, pruning, and knowledge distillation, and explain the trade-offs between model accuracy and inference latency.
Implementing Trustworthy AI
What to coverExplain how to incorporate explainability and auditability into AI systems. Discuss methods such as SHAP or LIME for model interpretability and how to address privacy concerns when training models on sensitive industry data.
Physics-Informed Neural Networks (PINNs)
What to coverDescribe the architecture and advantages of PINNs compared to traditional data-driven models. Focus on how incorporating physical laws as constraints in the loss function improves generalization, particularly in engineering or manufacturing applications.
MLOps and Pipeline Automation
What to coverDetail the components of a robust MLOps pipeline for industry applications. Address version control for data and models, automated testing, continuous integration/deployment (CI/CD) for AI, and monitoring for model drift in production environments.
Large Language Models and Multimodal Integration
What to coverDiscuss the challenges of integrating LLMs into specialized industry workflows. Address strategies for fine-tuning, RAG (Retrieval-Augmented Generation) for domain-specific knowledge, and the technical requirements for handling multimodal inputs.
Time-Series Modelling and Anomaly Detection
What to coverExplain approaches for detecting anomalies in high-frequency industrial sensor data. Compare statistical methods with deep learning architectures like LSTMs, GRUs, or Transformers, and discuss how to handle non-stationary data distributions.
Software Architecture for AI Solutions
What to coverDescribe how to design scalable software architectures that integrate AI models with existing industrial systems. Focus on the choice of programming languages (Python, C++, Java) based on performance requirements and the use of frameworks like Spark ML for big data processing.
Bayesian Methods and Uncertainty Quantification
What to coverExplain the importance of uncertainty quantification in high-stakes domains like healthcare or safety-critical engineering. Discuss how Bayesian neural networks or Monte Carlo dropout can provide confidence intervals for model predictions.
Behavioural practice
Translating Industry Requirements to Technical Proposals
What to coverDescribe a situation where you had to translate a non-technical partner's business problem into a concrete AI research proposal. Focus on how you clarified requirements, managed expectations, and defined success metrics.
Collaborating Across Diverse Stakeholders
What to coverProvide an example of working with a multidisciplinary team, such as PIs, industry partners, or other IHLs. Explain how you navigated conflicting priorities or technical disagreements to achieve a project outcome.
Managing Research Innovation and IP Creation
What to coverDescribe a project where you were tasked with creating intellectual property or innovative solutions from research. Explain your process for documenting findings and ensuring the work had potential for multiple adopters.
Adapting to Dynamic Research Environments
What to coverGive an example of how you handled a significant shift in project scope or technical direction. Detail the actions you took to maintain progress and how you communicated these changes to your team or stakeholders.