Principal Applied Scientist, Hardware Silicon and Systems Group

Amazon; Sunnyvale (CA); On-site; USD 130,000 - 190,000; Full time

Job summary

An innovative firm is seeking a Principal Applied Scientist to lead the development of cutting-edge ML models for consumer devices. In this pivotal role, you'll work at the intersection of machine learning and hardware optimization, shaping the future of AI in devices used by millions. Your expertise will drive the creation of novel architectures and optimization strategies, ensuring efficient on-device AI capabilities. Join a dynamic team dedicated to enhancing customer experiences through advanced technology and be part of a transformative journey in the consumer AI market.

Qualifications

• 8+ years of experience in machine learning with a focus on model architecture.
• Expertise in hardware-aware quantization and model compression techniques.

Responsibilities

• Own the technical architecture and optimization strategy for ML models.
• Develop novel model architectures optimized for custom silicon.

Skills

Machine Learning

Model Architecture Design

Model Optimization

Computer Architecture

Hardware Acceleration

Efficient Inference Algorithms

Quantization Techniques

Deep Learning Frameworks

Education

PhD in Computer Science

Master's in Electrical Engineering

Tools

TensorFlow

PyTorch

ONNX

Job description

Principal Applied Scientist, Hardware Silicon and Systems Group

Job ID: 2910791 | Amazon.com Services LLC

Our team leads the development and optimization of on-device ML models for Amazon's hardware products, including audio, vision, and multi-modal AI features. We work at the critical intersection of ML innovation and silicon design, ensuring AI capabilities can run efficiently on resource-constrained devices.

Currently, we enable production ML models across multiple device families, including Echo, Ring/Blink, and other consumer devices. Our work directly impacts Amazon's customer experiences in the consumer AI device market. The solutions we develop determine which AI features can be offered on-device versus requiring cloud connectivity, ultimately shaping product capabilities and customer experience across Amazon's hardware portfolio.

This is a unique opportunity to shape the future of AI in consumer devices at unprecedented scale. You'll be at the forefront of developing industry-first model architectures and compression techniques that will power AI features across millions of Amazon devices worldwide. Your innovations will directly enable new AI features that enhance how customers interact with Amazon products every day. Come join our team!

Key Job Responsibilities

As a Principal Applied Scientist, you will:

1. Own the technical architecture and optimization strategy for ML models deployed across Amazon's device ecosystem, from existing to yet-to-be-shipped products.
2. Develop novel model architectures optimized for our custom silicon, establishing new methodologies for model compression and quantization.
3. Create an evaluation framework for model efficiency and implement multimodal optimization techniques that work across vision, language, and audio tasks.
4. Define technical standards for model deployment and drive research initiatives in model efficiency to guide future silicon designs.
5. Spend the majority of your time doing deep technical work - developing novel ML architectures, writing critical optimization code, and creating proof-of-concept implementations that demonstrate breakthrough efficiency gains.
6. Influence architecture decisions impacting future silicon generations, establish standards for model optimization, and mentor others in advanced ML techniques.

BASIC QUALIFICATIONS

This role requires a blend of expertise at the intersection of ML and hardware optimization. You must be an expert in model training, with deep knowledge of cutting-edge architectures for vision, language, and multimodal tasks. Crucially, you need to be a specialist in hardware-aware quantization, with hands-on experience in model compression techniques like pruning and distillation. A strong background in computer architecture and familiarity with ML accelerator designs is essential, as is expertise in efficient inference algorithms and low-precision arithmetic.
Basic Qualifications:

1. Advanced degree (PhD preferred) in Computer Science, Electrical Engineering, or a related technical field.
2. 8+ years of experience in machine learning, with a focus on model architecture design, optimization, and deployment.
3. Expertise in developing and deploying deep learning models for real-world applications, including vision, language, and multimodal tasks.
4. Strong background in computer architecture, hardware acceleration, and efficient inference algorithms.
5. Hands-on experience with model compression techniques such as pruning, quantization, and distillation.
6. Proficiency with deep learning frameworks like TensorFlow, PyTorch, or ONNX.

PREFERRED QUALIFICATIONS

1. PhD in Computer Science, Electrical Engineering, or a related technical field.
2. 10+ years of experience in machine learning, with a track record of developing novel model architectures and optimization techniques.
3. Proven expertise in co-designing ML models and hardware accelerators for efficient on-device inference.
4. In-depth understanding of the latest advancements in model compression, including techniques like knowledge distillation, network pruning, and hardware-aware quantization.
5. Experience working on resource-constrained embedded systems and deploying ML models on edge devices.
6. Demonstrated ability to influence technical strategy and mentor cross-functional teams.
7. Strong communication skills and the ability to effectively present complex technical concepts to both technical and non-technical stakeholders.

Amazon is committed to a diverse and inclusive workplace. Amazon is an equal opportunity employer and does not discriminate on the basis of race, national origin, gender, gender identity, sexual orientation, protected veteran status, disability, age, or other legally protected status.