How our scientists connect and innovate at the Amazon Machine Learning Conference
To read more stories about our work and people at Amazon Ads, head back to our Careers homepage.
Every year more than 1,440 scientists, engineers, and product managers across Amazon gather to discuss everything from ethical AI to the quirks of recommendation models at our internal Amazon Machine Learning Conference (AMLC).
An academic community
“It feels like a proper academic conference,” says Neeti Narayan, a Senior Applied Scientist in Amazon Ads. “You’ve got the papers, the posters, the keynotes, but what makes AMLC different is who’s in the room. It’s not just scientists; it’s engineers, product managers, data scientists from all over the business. It’s a really collaborative environment.”
Neeti Narayan, Senior Applied Scientist at Amazon Ads
Neeti has been at Amazon for almost four years, after completing a PhD in deep learning (University at Buffalo: The State University of New York) and working for three years on natural language processing at a large tech company. Her day job involves developing AI-driven solutions that map advertiser demand to contextually relevant webpages, optimizing how businesses reach customers through Amazon Ads and product recommendations. Behind the scenes, this means training large language models to understand webpage context and product relevance, turning unstructured text into actionable signals.
That mix of theory and application is exactly why Neeti has thrown herself into supporting AMLC. She’s reviewed and published papers, and even organized workshops on how generative AI can be used in advertising. “I enjoy academic work,” she says. “Writing research papers, sharing them with a broader audience, getting feedback, it’s not just about documenting your work, it’s about improving it. And AMLC gives you that.”
Sharing ideas
For Martin Radfar, a Senior Research Scientist, AMLC is equally exciting. Martin has been with Amazon for six years, initially working on AI for Alexa before moving to Amazon Ads to explore image and video processing. His publication work is prolific, with 16 external publications in top AI conferences, but he still finds value in Amazon’s internal conference.
Martin Radfar, Senior Research Scientist at Amazon Ads
“In 2024, my paper was selected for oral presentation at the conference,” Martin explains. “Only about 10 percent make it, so it was a real privilege. Afterwards, people from other teams came up and said, ‘We could use part of your model.’ Helping other teams progress their work is very exciting. And later I found out mine was the second most downloaded paper from the whole conference.”
Martin’s research focuses on reversioning adverts, which means taking a creative asset and automatically transforming it for different platforms. “Each ad has components, product, logo, text. Different platforms need different formats,” he adds. “We built a system that can segment, rearrange, even generate a new background. It means advertisers can do in one click what used to take a whole design team.”
A culture of collaboration
The appeal of AMLC is partly intellectual, partly social. For Neeti, it was a paper on product reviews that stuck in her mind. “I reviewed it before the model went into production, about how to generate summaries of thousands of reviews using large language models,” she recalls. “Later we used their dataset in one of our own projects. That collaboration only happened because of AMLC.”
Martin has a similar story. A model his team developed to detect safe areas in an image for overlaying text caught the eye of another group of scientists, who then incorporated it into their own workflow. “It opens doors,” he says. “You don’t always realize your work has wider applications until someone else sees it.”
Investing in scientists
Amazon has been running AMLC since 2013. In 2024, there were 918 submissions, 89 made it to oral presentation, 190 to poster presentation. These are big numbers, but the impact is less about scale than connection. According to post-conference surveys, 89% of attendees left feeling more plugged into the science community at Amazon, a sign of the thriving scientific culture AMLC has fostered for over more than a decade.
This sense of belonging matters, especially in a company the size of Amazon. “Day to day, you’re focused on your own team, your own deadlines,” Neeti says. “But AMLC reminds you that you’re part of something much bigger. Suddenly you’re sitting next to someone from Alexa or AWS and realizing you’re tackling similar problems from different angles.”
The comparisons with external conferences are inevitable, and deliberate. “I’ve attended the big external ML conferences,” Neeti continues. “Honestly, the quality of the papers at AMLC is on par. The difference is that here you also see the immediate applications of the ideas presented. And you meet people who can help you implement these ideas into your own work.”
Martin agrees, though he’s quick to note that the pace of the field is changing how research is shared. “These days, many big tech companies are publishing less externally. Data privacy, intellectual property, it slows things down. But at AMLC, you can share work quickly and have a positive impact on your community.”
Impacting the future of science at Amazon
AMLC is also a barometer of where the field is heading. Last year’s keynotes included Andrew Ng, the Stanford professor and AI pioneer, who spoke about multi-agent systems, a topic Martin presented on. “I had just talked about multi-agent AI,” he laughs, “then Andrew Ng is on stage saying it’s the future of AI research. That was a nice moment.”
For Neeti, the inspiration is more in the sheer pace of discovery. “Generative AI is moving so fast it’s hard to keep up,” she says. “Teams are coming out with new tools every week. That pushes us to think bigger about ads, how campaigns are created, how products appear in Rufus, Amazon’s generative AI-powered shopping assistant. You see these ideas at AMLC and you start imagining the future.”
Both scientists are keen to explain that training large models, fine-tuning parameters, iterating until things work can be painstaking, but AMLC helps give this work a wider context. “You spend months working on a project,” says Neeti. “To then present it to thousands of colleagues, get their questions, their ideas, it’s energizing.”
Martin agrees, “as scientists we’re always balancing curiosity with impact. AMLC is where those two things meet.”
Breaking ad barriers
“We're at the forefront of innovation,” Alexis says. Being part of a team of consultants rather than traditional salespeople, Alexis is able to help her client solve real-world business problems. "When clients come to us with a challenge, we work backwards from that problem to find creative solutions."
But it’s not just about the tech; Alexis emphasizes the importance of Amazon’s culture of earning trust and having ownership. “I get to run my business how I see fit,” she says. “My bosses don’t micromanage but are there when I need them. This trust empowers me to think creatively and move quickly.”
Her innovative approach has earned recognition. One of her campaigns recently won a WPP global recognition award for work on activations that re-imagined creativity in an advertiser’s campaigns.