Machine Learning Intern

Integral Ad Science (IAS) is a global technology and data company that builds verification, optimization, and analytics solutions for the advertising industry and we’re looking for an AI intern to join our Data Science team. If you are excited by technology that has the power to handle hundreds of thousands of transactions per second; collect tens of billions of events each day; and evaluate thousands of data-points in real-time all while responding in just a few milliseconds, then IAS is the place for you!

We are looking for two AI Interns to join our Data Science team. These roles provide a unique opportunity to work on two exciting projects: Multimodal Model Evaluation and Data Curation for Multimodal Model Training.  If you are passionate about AI, data, and optimization, and want to gain hands-on experience with cutting-edge techniques, this internship is for you!

Project 1 – Multimodal Model Evaluation

In this project, you will explore and implement novel evaluation techniques for multimodal AI models. Your work will focus on defining objective benchmarks, analyzing models behavior across modalities (audio, text, images), and developing robust performance assessment strategies. This will ensure better reliability, fairness, and efficiency of multimodal AI systems for IAS use cases.

What you’ll get to do:

• Conduct an in-depth review of state-of-the-art multimodal model evaluation techniques.
• Identify and define key performance metrics to assess multimodal model effectiveness.
• Implement and test evaluation datasets across multiple modalities (text, audio, image, video) and their combinations.
• Explore zero-shot and few-shot learning approaches to leverage multimodal models for downstream tasks.
• Develop automated evaluation pipelines to streamline model assessment.
• Collaborate with cross-functional teams to integrate new evaluation methods into production systems.
• Analyze and report on evaluation results, providing insights for model improvements.

Project 2 – Data Curation for Model Training

This project focuses on curating and optimizing data to enhance training speed and performance for foundational models, particularly image encoders and multimodal models. Your efforts will directly impact the efficiency and effectiveness of model training processes.

What you’ll get to do:

• Focus on data quality improvement methods to accelerate model training and improve model performance.
• Hands-on experience with multimodal data (text, audio, image, video).
• Review and apply insights from state-of-the-art techniques like DINO v2 to optimize data for training.
• Conduct experiments to test data curation techniques and validate their impact on training times.
• Work closely with engineers to incorporate findings into model pipelines and training processes.
• Develop best practices for data handling and preprocessing to ensure high-quality inputs for model training.

You should apply if you have most of this experience: 

• Currently pursuing or recently completed a degree in Computer Science, Data Science, AI, or a related field.
• Experience with machine learning frameworks (e.g., PyTorch, TensorFlow).
• Familiarity with multimodal models.
• A proactive, problem-solving attitude with the ability to work independently.
  
• Project 1:
  • Familiarity with model evaluation techniques and performance metrics.
• Bonus points for:
  • Familiarity with current SOTA multimodal AI models.
  • Experience with large-scale data evaluation and testing.
• Project 2:
  • Strong understanding of data preprocessing, image data handling, and model training.
  • Familiarity with modern data quality improvement methods and their applications.
• Bonus points for:
  • Experience with advanced data augmentation and curation methods.
  • Familiarity with foundational models like Vision Transformers or other image encoders.

About Integral Ad Science

Integral Ad Science (IAS) is a leading global media measurement and optimization platform that delivers the industry’s most actionable data to drive superior results for the world’s largest advertisers, publishers, and media platforms. IAS’s software provides comprehensive and enriched data that ensures ads are seen by real people in safe and suitable environments, while improving return on ad spend for advertisers and yield for publishers. Our mission is to be the global benchmark for trust and transparency in digital media quality. For more information, visit integralads.com.

Equal Opportunity Employer:

IAS is an equal opportunity employer, committed to our diversity and inclusiveness. We will consider all qualified applicants without regard to race, color, nationality, gender, gender identity or expression, sexual orientation, religion, disability or age. We strongly encourage women, people of color, members of the LGBTQIA community, people with disabilities and veterans to apply.

California Applicant Pre-Collection Notice:

We collect personal information (PI) from you in connection with your application for employment or engagement with IAS, including the following categories of PI: identifiers, personal records, commercial information, professional or employment or engagement information, non-public education records, and inferences drawn from your PI. We collect your PI for our purposes, including performing services and operations related to your potential employment or engagement. For additional details or if you have questions, contact us at [email protected].

To learn more about us, please visit http://integralads.com/ 

Attention agency/3rd party recruiters: IAS does not accept any unsolicited resumes or candidate profiles. If you are interested in becoming an IAS recruiting partner, please send an email introducing your company to [email protected]. We will get back to you if there's interest in a partnership.

#LI-Hybrid