Jasper is the purpose-built generative AI platform for marketing. Our mission is to elevate marketers with the power of AI, to achieve better outcomes through their workflows and open up new business opportunities. Jasper securely trains on brand and strategy, accelerates content and campaign production, and helps Marketers measure and optimize performance as part of an end-to-end solution.

Jasper has been recognized as "one of the Top 15 Most Innovative AI Companies of 2024" by Fast Company. We have nearly 100k marketing team users with a growing roster of Enterprise and Fortune 500 logos including: Morningstar, Anthropologie, Prudential, Cushman & Wakefield, Wayfair, and more. Our teams are building multi-modal AI and working with the top AI platforms. We were one of the first partners of both OpenAI and Anthropic. With the Clipdrop acquisition, Jasper is redefining visual marketing across all modalities, offering advanced image features like background replacement, image upscaling, product staging, and more to our customers. We are transforming how marketers create, innovate, and captivate audiences.

We value being customer-obsessed and helping each other achieve the best outcomes through collaboration. We’re not just building solutions, we’re creating experiences that captivate and amaze us. We strive to deliver moments of wonder and delight that enable our customers to be more effective and creative in ways they never thought possible. Learn more at jasper.ai.

About the Role

Jasper Research is seeking a highly motivated intern to advance the frontiers of image-editing applications. In this role, you will be instrumental in developing state-of-the-art image models while collaborating closely with our talented team of researchers and engineers. The internship duration is 4 to 6 months, with a preference for a 6-month commitment.

Context

Diffusion Models (DM) have proven to be one of the most efficient class of generative models for many tasks ranging from image synthesis, video generation, audio or 3D applications. They have raised particular interest and enthusiasm for text-to-image applications where they outperform other approaches. However, their usability for real-time applications remains limited by the intrinsic iterative nature of their sampling mechanism. At inference time, these models aim at iteratively denoising a sample drawn from a Gaussian distribution to finally create a sample belonging to the data distribution. Nonetheless, such a denoising process requires multiple evaluations of a potentially very computationally costly neural function such that their faster counterparts such as Generative Adversarial Networks (GANs) and Variational AutoEncoders (VAEs) still tend to be preferred.

Goal of the internship

The main goal of this internship is to explore ideas to improve the sampling speed of diffusion models by exploring diverse recent ideas such as rectified flows or consistency models that allowed to reduce the number of sampling steps requires to sample efficiently from a trained diffusion model. This internship is a research internship the end goal of which is to write a research paper and share open-source if enough material

We have currently identified two potential starting paths for the internship:

1. Distillation-based methods that aims at extracting the knowledge of a pre-trained teacher model with a student model. In this approach the student is trained to predict the output of the teacher but in fewer steps

2. Simulation-free techniques such as rectified flows that can learn straighter ODE trajectories leading to potentially less approximation errors when solving the ODE (at sampling time).

Required Qualifications

• Currently enrolled in a Ph.D. or M.Sc program in Machine Learning, applied mathematics or computer science.

• A strong interest in the field, with a potentially proven track record through personal projects or previous experience in image generation.

• Being available for period of 4 to 6 months (6 months preferred)

• The ideal candidate will possess a strong critical thinking and problem-solving mindset, coupled with excellent teamwork skills.

Nice to have

• Strong programming Python skills, including software engineering best practices to produce high-quality code.

• Experience with distributed training and large-scale systems with PyTorch

• Proven track record of achieving significant results as demonstrated by first-authored publications in major conferences and journals such as CVPR, ECCV, ICCV, ICLR, NeurIPS

What We Offer

• Flexible remote work options with an office in central Paris

• Competitive compensation package

Our Culture

We value practical problem-solving abilities and a genuine passion for data. We're looking for data engineers who can transform raw data into actionable insights that drive business success. While we maintain high standards, we encourage applications from candidates who may not meet all criteria but demonstrate strong potential and a commitment to continuous learning.

Our goal is to be a diverse workforce that is representative at all job levels as we know the more inclusive we are, the better our product will be. We are committed to celebrating and supporting our differences and that diversity is essential to innovation and makes us better able to serve our customers. We hire people of all levels and backgrounds who are excited to learn and develop their skills.

We are an equal opportunity employer. Applicants will not be discriminated against because of race, color, creed, sex, sexual orientation, gender identity or expression, age, religion, national origin, citizenship status, disability, ancestry, marital status, veteran status, medical condition, or any protected category prohibited by local, state or federal laws.

By submitting this application, you acknowledge that you have reviewed and agree to Jasper's CCPA Notice to Candidates, available at legal.jasper.ai/#ccpa.