A5028665299- AI in cancer detection
- Domain Adaptation and Few-Shot Learning
- Radiomics and Machine Learning in Medical Imaging
- Artificial Intelligence in Healthcare and Education
- Multimodal Machine Learning Applications
- COVID-19 diagnosis using AI
- Clinical Laboratory Practices and Quality Control
- Distributed systems and fault tolerance
- Hematological disorders and diagnostics
- Cell Image Analysis Techniques
- Intelligent Tutoring Systems and Adaptive Learning
- Face Recognition and Perception
- Auction Theory and Applications
- Adversarial Robustness in Machine Learning
- Digital Imaging for Blood Diseases
- Anomaly Detection Techniques and Applications
- Optimization and Search Problems
- Biomedical Text Mining and Ontologies
- Explainable Artificial Intelligence (XAI)
- Data Visualization and Analytics
- Zoonotic diseases and public health
- Tuberculosis Research and Epidemiology
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
Berlin Institute for the Foundations of Learning and Data
2023-2024
Technische Universität Berlin
2023-2024
Technische Universität Braunschweig
2019
Histopathology plays a central role in clinical medicine and biomedical research. While artificial intelligence shows promising results on many pathological tasks, generalization dealing with rare diseases, where training data is scarce, remains challenge. Distilling knowledge from unlabelled into foundation model before learning from, potentially limited, labelled provides viable path to address these challenges. In this work, we extend the state of art models for digital pathology whole...
Recent advances in digital pathology have demonstrated the effectiveness of foundation models across diverse applications. In this report, we present a novel vision model based on RudolfV approach. Our was trained dataset comprising 1.2 million histopathology whole slide images, collected from two medical institutions: Mayo Clinic and Charit\'e - Universt\"atsmedizin Berlin. Comprehensive evaluations show that our achieves state-of-the-art performance twenty-one public benchmark datasets,...
BackgroundWhile previous studies of artificial intelligence (AI) have shown its potential for diagnosing diseases using imaging data, clinical implementation lags behind. AI models require training with large numbers examples, which are only available common diseases. In reality, however, the majority less frequent, and current overlook or misclassify them. An effective, comprehensive technique is needed full spectrum real-world diagnoses.MethodsWe collected two datasets gastrointestinal...
Today's computer vision models achieve human or near-human level performance across a wide variety of tasks. However, their architectures, data, and learning algorithms differ in numerous ways from those that give rise to vision. In this paper, we investigate the factors affect alignment between representations learned by neural networks mental inferred behavioral responses. We find model scale architecture have essentially no effect on with responses, whereas training dataset objective...
Multiple instance learning (MIL) is an effective and widely used approach for weakly supervised machine learning. In histopathology, MIL models have achieved remarkable success in tasks like tumor detection, biomarker prediction, outcome prognostication. However, explanation methods are still lagging behind, as they limited to small bag sizes or disregard interactions. We revisit through the lens of explainable AI (XAI) introduce xMIL, a refined framework with more general assumptions....
Deep learning has led to remarkable advancements in computational histopathology, e.g., diagnostics, biomarker prediction, and outcome prognosis. Yet, the lack of annotated data impact batch effects, systematic technical differences across hospitals, hamper model robustness generalization. Recent histopathological foundation models -- pretrained on millions billions images have been reported improve generalization performances various downstream tasks. However, it not systematically assessed...
This paper presents Contrastive Reconstruction, ConRec - a self-supervised learning algorithm that obtains image representations by jointly optimizing contrastive and self-reconstruction loss. We showcase state-of-the-art methods (e.g. SimCLR) have shortcomings to capture fine-grained visual features in their representations. extends the SimCLR framework adding (1) task (2) an attention mechanism within task. is accomplished applying simple encoder-decoder architecture with two heads. show...
Deep neural networks have reached human-level performance on many computer vision tasks. However, the objectives used to train these enforce only that similar images are embedded at locations in representation space, and do not directly constrain global structure of resulting space. Here, we explore impact supervising this by linearly aligning it with human similarity judgments. We find a naive approach leads large changes local representational harm downstream performance. Thus, propose...
We consider a central challenge that is mission critical for the successful operation of large-scale satellite constellations in Low-Earth Orbit: How can we coordinate short-term download operations enormous amounts generated data, based on wireless line-of-sight connections to limited number stationary ground station? These issues are future growth space systems, with multiple commercial operators competing downloading their data timely fashion, relying services scarce set stations subject...
Mit den Entwicklungen der Präzisionsmedizin steigen die Anforderungen an pathologische Diagnostik, histomorphologische und molekularpathologische Daten standardisiert, quantitativ integriert zu beurteilen. Große Hoffnungen werden in Verfahren Künstlichen Intelligenz (KI) gesetzt, gezeigt haben, komplexe klinische, histologische molekulare zur Krankheitsklassifikation, Biomarkerquantifizierung Prognoseabschätzung auswerten können. Diese Arbeit gibt einen Überblick über neueste KI Pathologie,...
While previous studies have demonstrated the potential of AI to diagnose diseases in imaging data, clinical implementation is still lagging behind. This partly because models require training with large numbers examples only available for common diseases. In reality, however, few are common, whereas majority less frequent (long-tail distribution). Current overlook or misclassify these We propose a deep anomaly detection approach that requires data from detect also all collected two...
Unsupervised learning has become an essential building block of AI systems. The representations it produces, e.g. in foundation models, are critical to a wide variety downstream applications. It is therefore important carefully examine unsupervised models ensure not only that they produce accurate predictions, but also these predictions "right for the wrong reasons", so-called Clever Hans (CH) effect. Using specially developed Explainable techniques, we show first time CH effects widespread...
The Platonic Representation Hypothesis claims that recent foundation models are converging to a shared representation space as function of their downstream task performance, irrespective the objectives and data modalities used train these models. Representational similarity is generally measured for individual datasets not necessarily consistent across datasets. Thus, one may wonder whether this convergence model representations confounded by commonly in machine learning. Here, we propose...
It is common practice to reuse models initially trained on different data increase downstream task performance. Especially in the computer vision domain, ImageNet-pretrained weights have been successfully used for various tasks. In this work, we investigate impact of transfer learning segmentation problems, being pixel-wise classification problems that can be tackled with encoder-decoder architectures. We find decoder does not help tasks, while encoder truly beneficial. demonstrate...