Staff Research Scientist, Artificial Intelligence
About the role
The Algorithmic Solutions Group develops cutting-edge, efficient algorithms to bring intelligence to the physical world. We fuse state-of-the-art machine learning with deep domain expertise to convert raw physical data into actionable insights, solving the hard problems where off-the-shelf solutions fall short.
This position challenges you to rethink AI breakthroughs, extending them beyond text and images to master complex physical signals—from multimodal sensory data to precision actuators and RF systems. You will architect and validate novel solutions that fuse modern AI with ADI technologies at the edge.
Key Responsibilities
Strategic Problem Definition: Collaborate with business leads and domain experts to identify opportunities where Modern AI can solve previously impossible problems. You will filter "hype" from "value," focusing on challenges that require deep technical innovation rather than off-the-shelf models.
Research-to-Product: Lead technical execution from mathematical conceptualization to proof-of-concept. You will partner with researchers and engineers across the organization to bridge the gap between abstract research papers and validated solutions.
Architecting Physical AI: Design next-generation neural architectures and custom training paradigms tailored to the physics of the data. You will investigate the inner workings of training dynamics and loss landscapes to develop robust learning strategies for complex physical signals.
Efficient AI: Bridge the gap between massive foundation models and edge constraints. You will research techniques in model distillation, optimization, and neural architecture search to deploy "Modern AI" on efficient compute platforms.
Thought Leadership: Maintain a deep awareness of the global AI research landscape. You will bring the best ideas from the academic community into ADI and mentor junior engineers.
The Ideal Candidate
You are a rigorous researcher and a pragmatic builder who thrives on complexity. You bring a "first-principles" understanding of deep learning, capable of deriving and modifying architectures from scratch.
You possess deep technical mastery of modern architectures such as Transformers, State Space Models (e.g., Mamba), and Diffusion Models. You are equally comfortable with advanced training paradims such as Self-Supervised Learning (SSL), Reinforcement Learning, and Flow Matching, or techniques for learning from limited data (few-shot/meta-learning).
You understand the mathematics behind these methods and can adapt them to novel modalities.
You combine theoretical depth with expert-level proficiency in PyTorch or JAX. You are experienced in developing, deploying, and optimizing models using modern frameworks and cloud platforms.
You navigate the ambiguity of early-stage innovation with creative persistence, translating open challenges into concrete technical roadmaps. You excel at decision-making under uncertainty, justifying how your architectural trade-offs directly address the problem and create value.
You distinguish yourself with: Edge Awareness: An understanding that our models must eventually leave the cloud. Experience with model quantization, distillation, or deploying to embedded targets is highly valued.
Familiarity with Circuits & Systems: Knowledge of signal chains, digital signal processing, and fundamental circuit concepts will allow you to bridge the gap between pure algorithms and the physical systems they control.
Fluency in "Signal": You are comfortable discussing Fourier transforms, noise floors, and sampling rates, and understanding how these concepts intersect with deep learning.
Strong publication record in top conferences and/or journals