Events & Conferences

Using supervised learning to train models for image clustering

Published

4 years ago

October 14, 2021

Most machine learning models use supervised learning, meaning they’re trained on annotated data, which is costly and time consuming to acquire.

The chief method for doing unsupervised learning, which doesn’t require annotated data, is clustering, or grouping data points together by salient characteristics. The idea is that each cluster represents some category, such as photos of the same person or the same species of animal.

To decide where to draw boundaries between clusters, clustering algorithms typically rely on heuristics, such as a threshold distance between cluster centers or the shape of the clusters’ distributions. In a paper we’re presenting at the International Conference on Computer Vision (ICCV), we propose, instead, to learn from data how to draw boundaries.

We first represent visual data using a graph, then use a graph neural network (GNN) to produce vector representations of the graph’s nodes. So far, we follow on previous work.

Instead of relying on heuristics, however, we use labeled data to learn how to cluster the vectors and, crucially, to decide how fine-grained those clusters should be. We call the labeled data meta-training data, since the goal is to learn a general clustering technique, not a specific classification model.

In particular, we propose a hierarchical GNN, meaning that it creates clusters by adding edges between nodes of a graph, then adds edges between the clusters to create still larger clusters, and so on, iterating until it decides that no more edges should be added.

A schematic of our graph-based hierarchical clustering approach. The colors of the image borders and of the graph nodes indicate data types (in this case, photos of the same actor). Our approach is hierarchical, iteratively treating small clusters generated at one level as the units of clustering for the next level. We call our base model LANDER, for link approximation and density estimation refinement, and our hierarchical clustering method Hi-LANDER.

Finally, we apply our hierarchical clustering technique to test sets whose classification categories are disjoint with those of the meta-training data. In our experiments we found that, compared to previous GNN-based supervised and unsupervised approaches, ours increased the F-score — which factors in both false positives and false negatives — by an average of 49% and 47%, respectively.

Constructing the graph

In our paper, we investigate the case in which we are training a model to cluster visual data that is similar to the meta-training data but has no class overlaps with it. For instance, the meta-training data might be faces of movie stars, while the target application is to cluster faces of politicians, athletes, or other public figures.

The first step in our process is to use the meta-training data to build a supervised classifier: if the meta-training data is faces of movie stars, the classifier labels input images with names of movie stars.

The classifier is an encoder-decoder model: the encoder produces a fixed-length vector representation of the input, or feature vector, and the decoder uses that vector to predict a label. Once we’ve trained the classifier, however, we use only the encoder for the rest of the process.

The feature vectors define points in a multidimensional space. On the basis of the vectors’ locations, we construct a graph, in which each node represents an image, and each image’s k nearest neighbors in the feature space are connected to it (share edges with it) in the graph.

This graph will serve as the input to the clustering model, which is also an encoder-decoder model. The encoder is a GNN, which produces a vector representation of each node in the graph, based on that node’s feature vector and those of the nodes it’s connected to. Call this vector the node embedding.

The clustering model

We adopt a hierarchical approach to clustering. Based on the node embeddings, the clustering model predicts edges between nodes. A cluster is defined as a group of nodes each of which shares an edge with at least one other node in the group and none of which shares an edge with any node outside the group.

Note that the goal of the clustering model is not just to reproduce the nearest-neighbor graph but to link nodes that represent data of the same type. The nearest-neighbor linkages are useful for predicting clustering linkages, but they are not identical with them.

After the first pass through the data, we aggregate each cluster into a single, representative “supernode” and repeat the whole process. That is, we create edges between each supernode and its k nearest neighbors, pass the resulting graph through the same GNN, and predict edges based on the supernode embeddings. We repeat this process until the clustering model predicts no edges between nodes.

We train our clustering model on two different objectives. One is to correctly predict links between nodes, where a correct link is one that picks out two representatives of the same data type in the meta-training data (say, two photos of the same actor).

We also train the model to correctly predict the density of a given data type in a given graph neighborhood. That is, for each node, the model should predict the proportion of nearby neighbors of the same data type.

Past research on clustering has shown that factoring in data density improves results. Previously, however, link prediction and data density prediction were handled by separate models. By using a single model to jointly predict both, we significantly increase computational efficiency. We believe that the combination also contributes to our increase in accuracy.

The other novelty of our approach is that, because of our hierarchical processing scheme, we optimize clustering across the entire input graph. Previous approaches would first divide the graph into subgraphs, then perform inference within subgraphs. This prevents natural parallelization, which is runtime efficient, and limits the effectiveness of information flow through the graph. The full graph-wide processing is another reason for our model’s improved efficiency.

In experiments, we considered two different sets of meta-training data. One consisted of closeups of human faces, the other of images of particular animal species. We tested the model trained on human faces on two other datasets, whose data categories had zero or very little overlap with those of the meta-training set — 0% and less than 2%. We tested the model trained on animal species on a dataset of previously unseen species. Across both models and the three test sets, our average improvements over previous GNN-based clustering models and unsupervised clustering methods were 49% and 47%, respectively.

In ongoing work, we are investigating the possibility training a more general clustering model, whose performance at inference time will be more transferrable across different data types — accurately clustering both faces and animal species, for instance.

Acknowledgements: Tianjun Xiao, Yongxin Wang, Yuanjun Xiong, Wei Xia, David Wipf, Zhang Zheng, Stefano Soatto

Source link

Related Topics:Clustering Graph neural networks (GNNs)ICCV Machine learning Unsupervised learning

Up Next

Explicit control of GAN-generated synthetic images

Don't Miss

Two Amazon-affiliated economists awarded Nobel Prize

Yifan Xing

Click to comment

Events & Conferences

A New Ranking Framework for Better Notification Quality on Instagram

Published

1 day 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

1 day 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.

Events & Conferences

Revolutionizing warehouse automation with scientific simulation

Published

1 week 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.