Principal Research Engineer – Robustness‑Explainability Coupling

corpora-jobs-1778796293285-db9d41c6 - Frontier Development

Research EngineerPrincipal1 position

⚡

Why This Role is Different

Frontier Development Role

Lead the end‑to‑end design of a robust, uncertainty‑aware explainability system that can be deployed in safety‑critical, multi‑agent environments. Your work will set the standard for how explanations survive adversarial attacks and evolving data streams.

The Frontier Element

You will pioneer a joint loss formulation that aligns gradient spaces of predictions and explanations, a novel Bayesian counterfactual sampler that guarantees epistemic coverage, and a real‑time explanation‑drift engine that operates at scale—none of which exist in current commercial pipelines.

🔬

Project Context

Research Area

Integrated Adversarial Explainability Training (IAT), Uncertainty‑Aware Counterfactual Fine‑Tuning (UAC‑FT), Adaptive Explanation Drift Monitoring (AEDM)

From: Overfitting of Explainability Models to Benign Data

Why This Role is Critical

This role unifies the core adversarial‑robustness and uncertainty‑aware components of the chapter, ensuring that explanations remain faithful under attack, high‑variance counterfactuals, and real‑time drift. It requires deep expertise in adversarial ML, Bayesian inference, and online monitoring to translate the research into production‑grade systems.

What You Will Build

A joint training pipeline that optimizes prediction and explanation fidelity, a Bayesian counterfactual fine‑tuning module that samples from weight posteriors, and a live drift‑detection service that triggers retraining or surrogate fallback. The deliverables include a reproducible benchmark suite, a Docker‑based inference stack, and a monitoring dashboard with explanation‑stability metrics.

🛠

Key Responsibilities

Design and implement the IAT loss that jointly optimizes classification accuracy and explanation fidelity under FGSM/PGD perturbations.
Develop a Bayesian counterfactual fine‑tuning framework that samples from weight posteriors and applies the Delta method for uncertainty estimation.
Build an online drift‑detection service using SHAP‑based metrics and Isolation Forests, integrated with Prometheus/Grafana dashboards.
Create a reproducible benchmark suite (TriGuard, Attribution Drift Score) and publish results to the research community.
Collaborate with Platform Engineers to containerize the training and inference pipelines for Kubernetes deployment.

🎯

Required Skills & Experience

Technical Must-Haves

PyTorch/TensorFlow deep‑learning frameworks

Expert

Implement custom loss functions, adversarial training loops, and Bayesian sampling.

Adversarial attack libraries (Foolbox, CleverHans)

Advanced

Generate FGSM/PGD attacks for training and evaluation.

Bayesian neural networks and probabilistic programming (Pyro, TensorFlow Probability)

Expert

Sample weight posteriors and compute variance‑based counterfactuals.

Statistical inference (Delta method, credible intervals)

Advanced

Derive uncertainty bounds for counterfactual constraints.

Explainability tools (Grad‑CAM, SHAP, LIME)

Proficient

Evaluate explanation fidelity and drift.

Docker, Kubernetes, Prometheus, Grafana

Proficient

Deploy and monitor the inference stack.

Experience Requirements

5+ years in adversarial ML or robust deep learning research.
Published work on explainability or robustness in top venues (ICLR, NeurIPS, CVPR).
Hands‑on experience building production‑grade ML pipelines.

Education

PhD in Computer Science, Machine Learning, or related field with a focus on robustness or explainability.

⭐

Preferred Skills

Experience with federated learning frameworks (TensorFlow Federated, PySyft).
Knowledge of EU AI Act compliance and audit‑ready explanation generation.

🤝

You Will Thrive Here If...

Demonstrated ability to ship end‑to‑end systems from research to production.
Comfort leading cross‑functional teams with minimal hand‑offs.
Proactive experimentation mindset—willing to iterate fast and learn from failures.

📈

Impact & Growth

12-Month Impact

Within 12 months, deliver a production‑ready explainability system that reduces explanation drift below 5% under adversarial perturbations, publishes a benchmark paper, and powers a live multi‑agent deployment in a regulated domain (e.g., autonomous driving or finance).

Growth Opportunity

Lead a growing team of research engineers, expand the framework to multi‑modal agents, and shape the company’s global strategy for trustworthy AI.

Ready to Push the Boundaries?

If this sounds like the challenge you have been looking for, we want to hear from you. We value what you can build over where you have been.