Events & Conferences

Anton van den Hengel’s journey from intellectual property law to computer vision pioneer

Published

3 years ago

June 20, 2022

Anton van den Hengel, an international pioneer in computer vision and its many applications, departed the University of Adelaide in South Australia to join Amazon as director of applied science in April 2020. He is creating a new, world-class machine-learning hub in Adelaide and supporting Amazon’s business through the development and application of state-of-the-art computer vision and scalable machine learning.

Related content

Senior principal scientist Aleix M. Martinez on why computer vision research has only begun to scratch the surface.

In 2018, van den Hengel was the founding director of the Australian Institute for Machine Learning (AIML), Australia’s first institute dedicated to machine learning research. When he left to join Amazon, AIML was 140 people strong and near the top of the institutional world rankings in terms of computer vision research. He remains the part-time director of AIML’s new Centre for Augmented Reasoning, whose mission is to build core Artificial Intelligence (AI) capability in Australia.

Van den Hengel has authored more than 300 research papers, commercialized eight patents, and been chief investigator on research projects funded by many Fortune 500 companies.

But it could all have been so different. The young van den Hengel first got into computer science simply to support his efforts to become an intellectual property lawyer. In fact, he completed his law degree.

Amazon in Australia

Research teams in Adelaide are developing state-of-the-art, large-scale machine learning methods and applications involving terabytes of data. They work on applying ML, and particularly computer vision, to a wide spectrum of areas.

“I’d bought the suit, tie, and bright white shirt and was all ready to start my first day as an entry level lawyer,” he recalls. “Then, instead, I turned around and went straight back into the University of Adelaide. I spent the next couple of decades there.”

What followed was a master’s, then PhD in computer science and, ultimately, building up the University of Adelaide’s forerunner to AIML, the Australian Centre for Visual Technologies.

The chance to have an impact

What turned van den Hengel around was the chance to study computer vision.

“I saw the opportunity to engage with something that I realized was going to have incredible impact,” he says. Computer vision and its applications are everywhere today, but in the early 1990s, things were very different. “It’s hard to believe now but at the time there were maybe 1000 people in the world working on computer vision, at a time when there weren’t any digital cameras,” he reminisces. “Most papers in CV were at least half about how people had taken the images.”

[In the early 90s] there were maybe 1000 people in the world working on computer vision, at a time when there weren’t any digital cameras. Most papers in CV were at least half about how people had taken the images.

Van den Hengel understood that humans are primarily visual animals and he clearly saw the inevitability of computers using vision to sense, and ultimately interact with, the world. “But back then, having a computer that could actually either measure or impact upon the real world was virtually unbelievable,” he says.

Since then, he says, computer vision has transformed from a heavily mathematical field with 300 people at every conference who all knew each other, to conferences of many thousands of people and auditoria full of companies trying to attract staff and sell things.

“The economic value of computer vision has gone through the roof,” he says.

Computer vision is a fundamental technology, van den Hengel says, because it relates the real world to symbols. “Humans reason about things in terms of symbols, so ‘cat’, ‘sky’, ‘car’, ‘road’, and ‘fish’ are all symbols, right? Computer vision takes visual signals from the real world and relates those signals to symbols,” he says.

That’s been the critical missing piece of the puzzle. For decades it was predicted that by the year 2000 we would have robots doing the housework and many other ‘magical’ things, but we came up short because there’s an infinite variation of things out there in the real world and it’s much harder to get a computer to reason about our physical environment than anybody imagined.”

Looking for answers

This missing piece is tackled by a subfield of computer vision known as visual question answering (VQA). The idea is to enable computers not only to understand the content of an image (or video/livestream) in a more semantic, human-like way, but also to answer questions posed in natural language about that image. For example, “Where was this photo taken?”, “Does it look like the person on the picnic blanket is expecting someone?”, “What’s the color of the dog nearest the stop sign?”.

Van den Hengel is the world’s most-cited researcher in VQA by an enormous margin, with close to 22,000 citations.

Fireside chat: Anton van den Hengel and Simon Lucey

“I got into it very early because I saw it as a threshold change in the way that artificial intelligence works,” van den Hengel says. “What’s interesting about VQA is that you ask the question at run-time and need the answer immediately, so it needs to be very flexible, unlike current machine learning applications, which are often fixed, single-purpose solutions to specific problems.”

In other words, it needs to be closer to true artificial intelligence – often referred to as artificial general intelligence.

In that vein, imagine a robot that could follow natural-language instructions, based on a greater understanding of what it sees around itself. It’s a sci-fi dream, but for how much longer?

In 2018, using a vision-and-language process similar to VQA, Van den Hengel and a team of colleagues from across Australia developed a simulator that uses imagery taken from the inside of real buildings to teach virtual agents to successfully navigate using visually grounded instructions, such as: “Head upstairs and walk past the piano through an archway directly in front. Turn right when the hallway ends at pictures and table. Wait by the moose antlers hanging on the wall.” It is only a matter of time before we can talk to our self-driving cars in a similar manner when necessary, says van den Hengel.

The power of neural networks

Rapid developments in machine learning are behind the recent supercharging of computer vision research.

“In the last 10 years of computer vision, we have essentially trained deep-learning neural networks to replace all of these lovely computer-vision algorithms that we’d previously come up with for solving a whole bunch of problems,” he says. “In fact, neural networks are so much better at it, they went from being just an interesting solution to a puzzle to being a practical solution to some of the core challenges we face.”

While at the University of Adelaide, van den Hengel has applied advances in ML and computer vision to make the world better in a variety of ways. These include working with Adelaide-based medical technology company LBT Innovations in creating an automated pathology machine called APAS (Automated Plate Assessment System) Independence, which can screen and interpret high volumes of pathology plates.

“There’s a shortage of trained pathologists, partly because it’s not a lot of fun sitting all day doing chemistry and looking at samples. APAS does the drudge work of the visual inspection process,” he says. The device was FDA approved in 2019.

Beyond computer vision, van den Hengel is currently the chief investigator for the Australian National Health and Medical Research Council’s Centre of Research Excellence in Healthy Housing, which is using ML to help deliver better outcomes within the Australian housing system, not only in terms of housing, but also in terms of health.

“People who are homeless suffer diseases and injuries, which put them into hospital, and homelessness can see people spiral into a set of difficult conditions that are very expensive for society to address,” he says. “It’s actually cheaper to house somebody than to fix the impact of homelessness. So where can we intervene in the housing process in a way that benefits everybody and also saves money?”

Not all of van den Hengel’s work is quite so serious, however.

“The paper I’m most happy about but that gets the least recognition is one that tells you how to build real Lego models of objects in images,” he says. “It’s got brilliant maths in it; some of my favorite maths. And it incorporates gravity, structural considerations and, you know, fantastic maths.” And did he mention the maths?

Van den Hengel has even used ML to design an IPA beer.

“Collecting the data was a real trauma: we had to drink, and rate, a lot of beer,” he laments. He named the resulting ale The Rodney, in homage to the Australian AI researcher and roboticist Rodney Brooks, whose work resulted in the Roomba vacuum cleaner.

Joining Amazon

Always an advocate for Australia on the world stage, van den Hengel was keen to play a leading role in Amazon’s research push into the country. “It was a fantastic opportunity to start a new group in Australia for a company like Amazon.”

Typically, when academics transition to Amazon, they talk about the increase in pace from academia to industry. Van den Hengel bucks that trend.

“I was running a group with 140 people, trying to make enough money to pay them, keep the doors open, deliver on projects for tens of millions of dollars, doing PR, you name it,” he says. “Here, I’ve got about 25 world-class people with PhDs who work for me and 12 interns.”

Van den Hengel noted that Amazon is a results-focused environment. “At Amazon you are expected to deliver, but you do it with an engineering team and support systems all geared towards delivering customer benefit.”

So what is van den Hengel delivering on? A current project is applying visual inspection methods to help to make sure that Amazon customers get the best fresh produce possible.

I think the whole retail field is moving towards a better understanding of the nature of objects in the world and how humans relate to those objects, or products. And that’s something that computer vision is particularly well-placed to deliver.

“Visual inspection is a magnificent challenge and a core problem in computer vision,” he says,” and addressing it means we can make sure that when a customer receives a delivery of, say, tomatoes, they are as perfect as can be.”

Another key project involves using computer vision and ML to understand in a deeper way the hundreds of millions of items in the ever-changing Amazon catalogue. The catalogue has a trove of information, both in the word-based product descriptions and the images supplied by sellers.

“Making the most of the information contained in these two sources of information – which is essentially what humans do – is an interesting challenge, because it relies on the relationships between visual signals and symbols,” he explains, adding that cracking this challenge will help customers who are using Amazon search find the product that best matches their need “even if they’re not entirely sure how best to specify it themselves.”

Despite the considerable demands of managing a growing team, van den Hengel is determined to remain hands-on with his own research. “Amazon’s an innovative company, and really, truly innovating in a way that’s going to provide something of value to customers that nobody else can means that you need managers who deeply understand where the technology can go,” he says.

So where is the technology going?

“I think the whole retail field is moving towards a better understanding of the nature of objects in the world and how humans relate to those objects, or products,” he says. “And that’s something that computer vision is particularly well-placed to deliver.”

Source link

Up Next

Olga Moskvyak’s journey into the world of science

Don't Miss

Calculating the differential cost of code changes

Sean O'Neill

Click to comment

Events & Conferences

A New Ranking Framework for Better Notification Quality on Instagram

Published

4 days ago

September 2, 2025

Xian Sun

We’re sharing how Meta is applying machine learning (ML) and diversity algorithms to improve notification quality and user experience.
We’ve introduced a diversity-aware notification ranking framework to reduce uniformity and deliver a more varied and engaging mix of notifications.
This new framework reduces the volume of notifications and drives higher engagement rates through more diverse outreach.

Notifications are one of the most powerful tools for bringing people back to Instagram and enhancing engagement. Whether it’s a friend liking your photo, another close friend posting a story, or a suggestion for a reel you might enjoy, notifications help surface moments that matter in real time.

Instagram leverages machine learning (ML) models to decide who should get a notification, when to send it, and what content to include. These models are trained to optimize for user positive engagement such as click-through-rate (CTR) – the probability of a user clicking a notification – as well as other metrics like time spent.

However, while engagement-optimized models are effective at driving interactions, there’s a risk that they might overprioritize the product types and authors someone has previously engaged with. This can lead to overexposure to the same creators or the same product types while overlooking other valuable and diverse experiences.

This means people could miss out on content that would give them a more balanced, satisfying, and enriched experience. Over time, this can make notifications feel spammy and increase the likelihood that people will disable them altogether.

The real challenge lies in finding the right balance: How can we introduce meaningful diversity into the notification experience without sacrificing the personalization and relevance people on Instagram have come to expect?

To tackle this, we’ve introduced a diversity-aware notification ranking framework that helps deliver more diverse, better curated, and less repetitive notifications. This framework has significantly reduced daily notification volume while improving CTR. It also introduces several benefits:

The extensibility of incorporating customized soft penalty (demotion) logic for each dimension, enabling more adaptive and sophisticated diversity strategies.
The flexibility of tuning demotion strength across dimensions like content, author, and product type via adjustable weights.
The integration of balancing personalization and diversity, ensuring notifications remain both relevant and varied.

The Risks of Notifications without Diversity

The issue of overexposure in notifications often shows up in two major ways:

Overexposure to the same author: People might receive notifications that are mostly about the same friend. For example, if someone often interacts with content from a particular friend, the system may continue surfacing notifications from that person alone – ignoring other friends they also engage with. This can feel repetitive and one-dimensional, reducing the overall value of notifications.

Overexposure to the same product surface: People might mostly receive notifications from the same product surface such as Stories, even when Feed or Reels could provide value. For example, someone may be interested in both reel and story notifications but has recently interacted more often with stories. Because the system heavily prioritizes past engagement, it sends only story notifications, overlooking the person’s broader interests.

Introducing Instagram’s Diversity-Aware Notification Ranking Framework

Instagram’s diversity-aware notification ranking framework is designed to enhance the notification experience by balancing the predicted potential for user engagement with the need for content diversity. This framework introduces a diversity layer on top of the existing engagement ML models, applying multiplicative penalties to the candidate scores generated by these models, as figure1, below, shows.

The diversity layer evaluates each notification candidate’s similarity to recently sent notifications across multiple dimensions such as content, author, notification type, and product surface. It then applies carefully calibrated penalties—expressed as multiplicative demotion factors—to downrank candidates that are too similar or repetitive. The adjusted scores are used to re-rank the candidates, enabling the system to select notifications that maintain high engagement potential while introducing meaningful diversity. In the end, the quality bar selects the top-ranked candidate that passes both the ranking and diversity criteria.

Figure.1: Instagram’s diversity-aware ranking framework where the diversity layer sits on top of the existing modeling layer and penalizes notifications that are too similar to recently sent ones.

Mathematical Formulation

Within the diversity layer, we apply a multiplicative demotion factor to the base relevance score of each candidate. Given a notification candidate 𝑐, we compute its final score as the product of its base ranking score and a diversity demotion multiplier:

$\text{Score}(c) = R(c) \times D(c)$

where R(c) represents the candidate’s base relevance score, and D(c) ∈ [0,1] is a penalty factor that reduces the score based on similarity to recently sent notifications. We define a set of semantic dimensions (e.g., author, product type) along which we want to promote diversity. For each dimension i, we compute a similarity signal p_i(c) between candidate c and the set of historical notifications H, using a maximal marginal relevance (MMR) approach:

$p_i(c) = \mathrm{max}_{h \in H}\mathrm{sim}_i(c, h)$

where sim_i(·,·) is a predefined similarity function for dimension i. In our baseline implementation, p_i(c) is binary: it equals 1 if the similarity exceeds a threshold 𝜏_i and 0 otherwise.

The final demotion multiplier is defined as:

$D(c) = \prod_{i=1}^{m} \left( 1 - w_i \cdot p_i(c) \right)$

where each w_i∈ [0,1] controls the strength of demotion for its respective dimension. This formulation ensures that candidates similar to previously delivered notifications along one or more dimensions are proportionally down-weighted, reducing redundancy and promoting content variation. The use of a multiplicative penalty allows for flexible control across multiple dimensions, while still preserving high-relevance candidates.

The Future of Diversity-Aware Ranking

As we continue evolving our notification diversity-aware ranking system, a next step is to introduce more adaptive, dynamic demotion strategies. Instead of relying on static rules, we plan to make demotion strength responsive to notification volume and delivery timing. For example, as a user receives more notifications—especially of similar type or in rapid succession—the system progressively applies stronger penalties to new notification candidates, effectively mitigating overwhelming experiences caused by high notification volume or tightly spaced deliveries.

Longer term, we see an opportunity to bring large language models (LLMs) into the diversity pipeline. LLMs can help us go beyond surface-level rules by understanding semantic similarity between messages and rephrasing content in more varied, user-friendly ways. This would allow us to personalize notification experiences with richer language and improved relevance while maintaining diversity across topics, tone, and timing.

Source link

Events & Conferences

Simplifying book discovery with ML-powered visual autocomplete suggestions

Published

4 days ago

September 2, 2025

Mao Sheng Liu

Every day, millions of customers search for books in various formats (audiobooks, e-books, and physical books) across Amazon and Audible. Traditional keyword autocomplete suggestions, while helpful, usually require several steps before customers find their desired content. Audible took on the challenge of making book discovery more intuitive and personalized while reducing the number of steps to purchase.

We developed an instant visual autocomplete system that enhances the search experience across Amazon and Audible. As the user begins typing a query, our solution provides visual previews with book covers, enabling direct navigation to relevant landing pages instead of the search result page. It also delivers real-time personalized format recommendations and incorporates multiple searchable entities, such as book pages, author pages, and series pages.

1 of 2

Audible’s visual-autocomplete experience.

2 of 2

Amazon’s visual-autocomplete experience.

Our system needed to understand user intent from just a few keystrokes and determine the most relevant books to display, all while maintaining low latency for millions of queries. Using historical search data, we match keystrokes to products, transforming partial inputs into meaningful search suggestions. To ensure quality, we implemented confidence-based filtering mechanisms, which are particularly important for distinguishing between general queries like “mystery” and specific title searches. To reflect customers’ most recent interests, the system applies time-decay functions to long historical user interaction data.

Related content

Assessing the absolute utility of query results, rather than just their relative utility, improves learning-to-rank models.

To meet the unique requirements of each use case, we developed two distinct technical approaches. On Audible, we deployed a deep pairwise-learning-to-rank (DeepPLTR) model. The DeepPLTR model considers pairs of books and learns to assign a higher score to the one that better matches the customer query.

The DeepPLTR model’s architecture consists of three specialized towers. The left tower factors in contextual features and recent search patterns using a long-short-term-memory model, which processes data sequentially and considers its prior decisions when issuing a new term in the sequence. The middle tower handles keyword and item engagement history. The right tower factors in customer taste preferences and product descriptions to enable personalization. The model learns from paired examples, but at runtime, it relies on books’ absolute scores to assemble a ranked list.

Training architecture of the DeepPLTR model, which takes in paired examples (green and pink blocks). At runtime, the model scores only a single candidate at a time.

For Amazon, we implemented a two-stage modeling approach involving a probabilistic information-retrieval model to determine the book title that best matches each keyword and a second model that personalizes the book format (audiobooks, e-books, and physical books). This dual-strategy approach maintains low latency while still enabling personalization.

In practice, a customer who types “dungeon craw” in the search bar now sees a visual recommendation for the book Dungeon Crawler Carl, complete with book cover, reducing friction by bypassing a search results page and sending the customer directly to the product detail page. On Audible, the system also personalizes autocomplete results and enriches the discovery experience with relevant connections. These include links to the author’s complete works (Matt Dinniman’s author page) and, for titles that belong to a series, links to the full collection (such as the Dungeon Crawler Carl series).

Related content

Using reinforcement learning improves candidate selection and ranking for search, ad platforms, and recommender systems.

On Amazon, when the customer clicks on the title, the model personalizes the right book-format (audiobooks, e-books, physical books) recommendation and directs the customer to the right product detail page.

In both cases, after the customer has entered a certain number of keystrokes, the system employs a model to detect customer intent (e.g., book title intent for Amazon or author intent for Audible) and determine which visual widget should be displayed.

Audible and Amazon books’ visual autocomplete provides customers with more relevant content more rapidly than traditional autocomplete, and its direct navigation reduces the number of steps to find and access desired books — all while handling millions of queries at low latency.

This technology is not just about making book discovery easier; it is laying the foundation for future improvements in search personalization and visual discovery across Amazon’s ecosystem.

Acknowledgements: Jiun Kim, Sumit Khetan, Armen Stepanyan, Jack Xuan, Nathan Brothers, Eddie Chen, Vincent Lee, Soumy Ladha, Justine Luo, Yuchen Zeng, David Torres, Gali Deutsch, Chaitra Ramdas, Christopher Gomez, Sharmila Tamby, Melissa Ma, Cheng Luo, Jeffrey Jiang, Pavel Fedorov, Ronald Denaux, Aishwarya Vasanth, Azad Bajaj, Mary Heer, Adam Lowe, Jenny Wang, Cameron Cramer, Emmanuel Ankrah, Lydia Diaz, Suzette Islam, Fei Gu, Phil Weaver, Huan Xue, Kimmy Dai, Evangeline Yang, Chao Zhu, Anvy Tran, Jessica Wu, Xiaoxiong Huang, Jiushan Yang

Source link

Events & Conferences

Revolutionizing warehouse automation with scientific simulation

Published

2 weeks ago

August 26, 2025

Deniz Akyildiz

Modern warehouses rely on complex networks of sensors to enable safe and efficient operations. These sensors must detect everything from packages and containers to robots and vehicles, often in changing environments with varying lighting conditions. More important for Amazon, we need to be able to detect barcodes in an efficient way.

Related content

Generative AI supports the creation, at scale, of complex, realistic driving scenarios that can be directed to specific locations and environments.

The Amazon Robotics ID (ARID) team focuses on solving this problem. When we first started working on it, we faced a significant bottleneck: optimizing sensor placement required weeks or months of physical prototyping and real-world testing, severely limiting our ability to explore innovative solutions.

To transform this process, we developed Sensor Workbench (SWB), a sensor simulation platform built on NVIDIA’s Isaac Sim that combines parallel processing, physics-based sensor modeling, and high-fidelity 3-D environments. By providing virtual testing environments that mirror real-world conditions with unprecedented accuracy, SWB allows our teams to explore hundreds of configurations in the same amount of time it previously took to test just a few physical setups.

Camera and target selection/positioning

Sensor Workbench users can select different cameras and targets and position them in 3-D space to receive real-time feedback on barcode decodability.

Three key innovations enabled SWB: a specialized parallel-computing architecture that performs simulation tasks across the GPU; a custom CAD-to-OpenUSD (Universal Scene Description) pipeline; and the use of OpenUSD as the ground truth throughout the simulation process.