Applied Researcher I

Capital One Capital One · Banking · New York, NY +3

Applied Researcher I role focused on building AI foundation models through all phases of development, from design through training, evaluation, validation, and implementation. The role involves high-impact applied research to push the latest AI developments into next-generation customer experiences, leveraging technologies like Pytorch, AWS Ultraclusters, Huggingface, Lightning, and VectorDBs. Experience with training optimization, self-supervised learning, robustness, explainability, and RLHF is desired, with a track record of delivering models at scale.

What you'd actually do

  1. Partner with a cross-functional team of data scientists, software engineers, machine learning engineers and product managers to deliver AI-powered products that change how customers interact with their money.
  2. Leverage a broad stack of technologies — Pytorch, AWS Ultraclusters, Huggingface, Lightning, VectorDBs, and more — to reveal the insights hidden within huge volumes of numeric and textual data.
  3. Build AI foundation models through all phases of development, from design through training, evaluation, validation, and implementation.
  4. Engage in high impact applied research to take the latest AI developments and push them into the next generation of customer experiences.
  5. Flex your interpersonal skills to translate the complexity of your work into tangible business goals.

Skills

Required

  • PhD in Electrical Engineering, Computer Engineering, Computer Science, AI, Mathematics, or related fields, with an exception that required degree will be obtained on or before the scheduled start date or M.S. in Electrical Engineering, Computer Engineering, Computer Science, AI, Mathematics, or related fields plus 2 years of experience in Applied Research
  • hands-on experience developing AI foundation models and solutions using open-source tools and cloud computing platforms
  • deep understanding of the foundations of AI methodologies
  • experience building large deep learning models, whether on language, images, events, or graphs
  • expertise in one or more of the following: training optimization, self-supervised learning, robustness, explainability, RLHF
  • engineering mindset as shown by a track record of delivering models at scale both in terms of training data and inference volumes
  • experience in delivering libraries, platform level code or solution level code to existing products
  • track record of coming up with high quality ideas or improving upon existing ideas in machine learning, demonstrated by accomplishments such as first author publications or projects
  • ability to own and pursue a research agenda, including choosing impactful research problems and autonomously carrying out long-running projects

Nice to have

  • LLM
  • PhD focus on NLP or Masters with 5 years of industrial NLP research experience
  • Multiple publications on topics related to the pre-training of large language models (e.g. technical reports of pre-trained LLMs, SSL techniques, model pre-training optimization)
  • Member of team that has trained a large language model from scratch (10B + parameters, 500B+ tokens)
  • Publications in deep learning theory
  • Publications at ACL, NAACL and EMNLP, Neurips, ICML or ICLR
  • PhD focused on topics related to optimizing training of very large deep learning models
  • Multiple years of experience and/or publications on one of the following topics: Model Sparsification, Quantization, Training Parallelism/Partitioning Design, Gradient Checkpointing, Model Compression
  • Experience optimizing training for a 10B+ model
  • Deep knowledge of deep learning algorithmic and/or optimizer design
  • Experience with compiler design
  • PhD focused on topics related to guiding LLMs with further tasks (Supervised Finetuning, Instruction-Tuning, Dialogue-Finetuning, Parameter Tuning)
  • Demonstrated knowledge of principles of transfer learning, model adaptation and model guidance
  • Experience deploying a fine-tuned large language model

What the JD emphasized

  • track record of delivering models at scale
  • first author publications or projects
  • publications on topics related to the pre-training of large language models
  • trained a large language model from scratch
  • publications in deep learning theory
  • publications at ACL, NAACL and EMNLP, Neurips, ICML or ICLR
  • optimizing training of very large deep learning models
  • optimizing training for a 10B+ model
  • guiding LLMs with further tasks (Supervised Finetuning, Instruction-Tuning, Dialogue-Finetuning, Parameter Tuning)

Other signals

  • applied research
  • foundation models
  • training optimization
  • self-supervised learning
  • explainability
  • RLHF
  • delivering models at scale