Emma Beauxis-Aussalet

Vrije Universiteit Amsterdam | VU · Department of Computer Science

Classifiers can provide counts of items per class, but systematic classification errors yield biases (e.g., if a class is often misclassified as another, its size may be under-estimated). To handle classification biases, the statistics and epidemiology domains devised methods for estimating unbiased class sizes (or class probabilities) without iden...

Handling classification uncertainty is a crucial challenge for supporting efficient and ethical classification systems. This thesis addresses uncertainty issues from the perspective of end-users with limited expertise in machine learning. We focus on uncertainties that pertain to estimating class sizes, i.e., numbers of objects per class. We aim at...

Classifiers are applied in many domains where classification errors have significant implications. However, end-users may not always understand the errors and their impact, as error visualizations are typically designed for experts and for improving classifiers. We discuss the specific needs of classifiers' end-users and a simplified visualization,...

[Problem Introduction paper accepted at ICML 2019 workshop on AI for Social Good - https://aiforsocialgood.github.io/icml2019/index.htm] Abusive online behaviors occur at a large scale on all social media, and have dire consequences for their victims. Although the problem is largely acknowledged , technological solutions remain limited to detecting...

Keynote at Amsterdam Data Science

The increasing use of artificial intelligence (AI) technologies across application domains has prompted our society to pay closer attention to AI's trustworthiness, fairness, interpretability, and accountability. In order to foster trust in AI, it is important to consider the potential of interactive visualization, and how such visualizations help...

Machine learning systems can make more errors for certain populations and not others, and thus create discriminations. To assess such fairness issue, errors are typically compared across populations. We argue that we also need to account for the variability of errors in practice , as the errors measured in test data may not be exactly the same in r...

Recommender systems greatly impact the scope of information we are exposed to. For critical topics like the climate crisis, the filter bubbles and echo chambers of recommendation systems can shape our opinion, and perhaps our future. Recommendation systems can become instruments of social engineering, in ways we might not wish, nor control. We will...

Tips for future PhD students, presented at TECH020 an event of Amsterdam Data Science.

What is classification bias? How to measure it? How to visualize it? ...and how to make it understandable.

Classifiers are applied in many domains where classification errors have significant implications. However, end-users may not always understand the errors and their impact, as error visualizations are typically designed for experts and for improving classifiers. We discuss the specific needs of classifiers' end-users, and a simplified visualization...

Re-submission of the eponym paper published at DSAA 2017

We designed a system that supports eXplainable AI (XAI) by letting users explore the specific features that have led an AI to make a specific classification. Users can also explore how these specific features would impact the end-results if different forms of AI were used. Team members: Emma Beauxis-Aussalet, Dan Xu, Abdelrahman Hassan, Aysenur Bi...

Presentation of our ECCE 2018 paper

Classifiers are applied in many domains where classification errors have significant implications. However, end-users may not always understand the errors and their impact, as error visualizations are typically designed for experts and for improving classifiers. We discuss a visualization design that addresses the specific needs of classifiers' end...

Classifiers are applied in many domains where classification errors have significant implications. However, end-users may not always understand the errors and their impact, as error visualizations are typically designed for experts and for improving classifiers. We discuss a visualization design that addresses the specific needs of classifiers' end...

Most multi-class classifiers make their prediction for a test sample by scoring the classes and selecting the one with the highest score. Analyzing these prediction scores is useful to understand the classifier behavior and to assess its reliability. We present an interactive visualization that facilitates per-class analysis of these scores. Our sy...

Uncertainty impacts many crucial issues the world is facing today – from climate change prediction, to scientific modelling, to the interpretation of medical data. Decisions typically rely on data which can be aggregated from different sources and further transformed using a variety of algorithms and models. Such data processing pipelines involve d...

Uncertainty impacts many crucial issues the world is facing today – from climate change prediction, to scientific modelling, to the interpretation of medical data. Decisions typically rely on data which can be aggregated from different sources and further transformed using a variety of algorithms and models. Such data processing pipelines involve d...

Visualization components and online user interface for exploring classification errors. The D3 components provide both expert visualizations and simplified visualization for end-users who are not experts in machine learning.

Computer Vision is a promising technique for in-situ monitoring of ecosystems. It is non-intrusive and cost-effective compared to sending human observers. Automatic animal detection and species recognition support the study of population dynamics and species composition, i.e., the evolution of species populations' size. Fixed cameras support contin...

Computer vision technology has been considered in marine ecology research as a innovative, promising data collection method. It contrasts with traditional practices in the information that is collected, and its inherent errors and biases. Ecology research is based on the analysis of biological characteristics (e.g., species, size, age, distribution...

Several data analysis steps are required for understanding computer vision results and drawing conclusions about the actual trends in the fish populations. Particular attention must be drawn to the potential errors that can impact the scientific validity of end-results. This chapter discusses the means for ecologists to investigate the uncertainty...

Machine Learning classifiers can be used to analyze trends in counts of items per class, e.g., over time or location. This counting task is a basis for a variety of data analysis use cases, such as the study of species populations living in an ecosystem, or the profiling of customers using a service or a product. Classification results are inherent...

Machine Learning classifiers can be used to analyze trends in counts of items per class, e.g., over time or location. This counting task is a basis for a variety of data analysis use cases, such as the study of species populations living in an ecosystem, or the profiling of customers using a service or a product. Classification results are inherent...

Machine learning is widely used for mining collections, such as images, sounds, or texts, by classifying their elements into categories. Automatic classification based on supervised learning requires groundtruth datasets for modeling the elements to classify, and for testing the quality of the classification. Because collecting groundtruth is tedio...

Computer vision enables in-situ monitoring of animal populations at a lower cost and with less ecosystem disturbance than with human observers. However, computer vision uncertainty may not be fully understood by end-users, and the uncertainty assessments performed by technology experts may not fully address end-user needs. This knowledge gap can yi...

Video analysis tools can provide valuable datasets for a wide range of applications, such as monitoring animal populations for ecology research, while reducing human efforts for collecting information. Transferring such technology to novel application domains implies exposing non-expert users to unfamiliar datasets and technical concepts. Existing...

Machine Learning techniques for automatic classification have reached a broad range of applications. But the technology transfers face issues with user trust and acceptance, as classification results inherently contain errors. Machine Learning experts rely on widely-established error measurement methods and uncertainty visualizations. However end-u...

We introduce an online user interface for visualizing multidimensional data, with a simple and highly flexible interaction design. The multidimensional data of the demo describes fish populations. It is extracted from continuous video footage, collected from 9 fixed cameras that observed coral reef ecosystems during 3 years. Computer Vision algorit...

We present an interface for eliciting sets of acceptable gambles on a three-outcome possibility space, discuss an experiment conducted for testing this interface, and present the results of this experiment.

----------------------------------------------------------------------------------------------------------------------------------------- A more recent & complete publication is available: https://www.researchgate.net/publication/325958191_Supporting_End-User_Understanding_of_Classification_Errors -------------------------------------------------...

Machine Learning uncertainty is not easily understood by end-users. It yields trust issues impeding technology transfer, or potentially critical misinterpretations of data. The state-of-the-art visualizations are complex, and potentially misleading. Our novel design reduces complexity and limits misinterpretations of the uncertainty in classificati...

Applying computer vision for monitoring ecosystems opens new perspectives for marine ecology research. Information on fish populations can be automatically extracted from underwater cameras. As scientists, ecologists need to understand the uncertainty in computer vision results. We present an interface for studying fish populations that addresses t...

The recent use of computer vision techniques for monitoring ecosystems has opened new perspectives for marine ecology research. These techniques can extract information about fish populations from in-situ cameras, without requiring ecologists to watch the videos. However, they inherently introduce uncertainty since automatic information extraction...

In-situ video recording of underwater ecosystems is able to provide valuable information for biology research and natural resources management, e.g. changes in species abundance. Searching the videos manually, however, requires costly human effort. Our video analysis tool supports the key task of counting different species of fish, allowing marine...

In this work we present a framework for fish population monitoring through the analysis of underwater videos. We specifically focus on the user information needs, and on the dynamic data extraction and retrieval mechanisms that support them. Sophisticated though a software tool may be, it is ultimately important that its interface satisfies users'...

Understanding and analyzing fish behaviour is a fundamental task for biologists that study marine ecosystems because the changes in animal behaviour reflect environmental conditions such as pollution and climate change. To support investigators in addressing these complex questions, underwater cameras have been recently used. They can continuously...

Long-term monitoring of the underwater environment is still labour intensive work. Using underwater surveillance cameras to monitor this environment has the potential advantage to make the task become less labour intensive. Also, the obtained data can be stored making the research reproducible. In this work, a system to analyse long-term underwater...