Amit Sethi
YOU?
Author Swipe
View article: EXP-CAM: Explanation Generation and Circuit Discovery Using Classifier Activation Matching
EXP-CAM: Explanation Generation and Circuit Discovery Using Classifier Activation Matching Open
Machine learning models, by virtue of training, learn a large repertoire of decision rules for any given input, and any one of these may suffice to justify a prediction. However, in high-dimensional input spaces, such rules are difficult t…
View article: Mix, Align, Distil: Reliable Cross-Domain Atypical Mitosis Classification
Mix, Align, Distil: Reliable Cross-Domain Atypical Mitosis Classification Open
Atypical mitotic figures (AMFs) are important histopathological markers yet remain challenging to identify consistently, particularly under domain shift stemming from scanner, stain, and acquisition differences. We present a simple trainin…
View article: Enhancing stroke recovery assessment: A machine learning approach to real-world hand function analysis
Enhancing stroke recovery assessment: A machine learning approach to real-world hand function analysis Open
This study presents a machine learning framework that accurately classifies clinically relevant real-world upper limb function using accelerometer data. This approach offers a more precise and objective assessment of post-stroke hand use, …
View article: Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts
Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts Open
We propose a modular framework for predicting cancer specific survival directly from whole slide pathology images (WSIs). The framework consists of four key stages designed to capture prognostic and morphological heterogeneity. First, a Qu…
View article: Network Inversion for Uncertainty-Aware Out-of-Distribution Detection
Network Inversion for Uncertainty-Aware Out-of-Distribution Detection Open
Out-of-distribution (OOD) detection and uncertainty estimation (UE) are critical components for building safe machine learning systems, especially in real-world scenarios where unexpected inputs are inevitable. However the two problems hav…
View article: A Surrogate Model for the Forward Design of Multi-layered Metasurface-based Radar Absorbing Structures
A Surrogate Model for the Forward Design of Multi-layered Metasurface-based Radar Absorbing Structures Open
Metasurface-based radar absorbing structures (RAS) are highly preferred for applications like stealth technology, electromagnetic (EM) shielding, etc. due to their capability to achieve frequency selective absorption characteristics with m…
View article: Supplementary Table S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Table S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
4-part table showing hospital sources for cases in the training, validation, initial testing, and final testing subsets.
View article: Supplementary Figure S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Figure S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Figure S1. Scatterplot of LumA proportion by transcriptomic analysis versus percentage of tumor image patches classified as LumA by the DNN model - including held-out cases with nonLumA PAM50 assignment (n = 256). Best-fittin…
View article: Supplementary Table S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Table S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Table S2. Clinical and molecular features of PAM50 Luminal A breast cancers in the independent test set (n = 230) according to degree of adherence of transcriptomic profile to the LumA subtype by semi-supervised noon-negative…
View article: Supplementary Figure S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Figure S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Figure S2. Paired scatterplots comparing 17 Luminal A cases from the test set obtained from hospitals that did not contribute cases for model training versus the 213 Luminal A cases remaining in the test set. Correlations bet…
View article: Network Inversion of Convolutional Neural Nets (Student Abstract)
Network Inversion of Convolutional Neural Nets (Student Abstract) Open
Neural networks have emerged as powerful tools across various applications, yet their decision-making process often remains opaque, leading to them being perceived as "black boxes." This opacity raises concerns about their interpretability…
View article: WaveMixSR-V2: Enhancing Super-resolution with Higher Efficiency (Student Abstract)
WaveMixSR-V2: Enhancing Super-resolution with Higher Efficiency (Student Abstract) Open
Recent advancements in single image super-resolution have been predominantly driven by token mixers and transformer architectures. WaveMixSR utilized the WaveMix architecture, employing a two-dimensional discrete wavelet transform for spat…
View article: Shortcut Learning Susceptibility in Vision Classifiers
Shortcut Learning Susceptibility in Vision Classifiers Open
Shortcut learning, where machine learning models exploit spurious correlations in data instead of capturing meaningful features, poses a significant challenge to building robust and generalizable models. This phenomenon is prevalent across…
View article: Low-Resource Video Super-Resolution using Memory, Wavelets, and Deformable Convolutions
Low-Resource Video Super-Resolution using Memory, Wavelets, and Deformable Convolutions Open
The tradeoff between reconstruction quality and compute required for video super-resolution (VSR) remains a formidable challenge in its adoption for deployment on resource-constrained edge devices. While transformer-based VSR models have s…
View article: Privacy Preserving Properties of Vision Classifiers
Privacy Preserving Properties of Vision Classifiers Open
Vision classifiers are often trained on proprietary datasets containing sensitive information, yet the models themselves are frequently shared openly under the privacy-preserving assumption. Although these models are assumed to protect sen…
View article: Table 2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Table 2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
PFS for the test cohort of PAM50-assigned LumA breast cancer, comparing pure vs. admixed cases based on the proportion of the tumor image classified as LumA by the deep learning model
View article: Data from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Data from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Intratumor heterogeneity (ITH) presents challenges for precision oncology, but methods for its spatial quantification, scalable at population levels, do not exist. Based on previous work showing that the admixture of PAM50 subtype can be m…
View article: Table 1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Table 1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Clinical and molecular features of PAM50 LumA breast cancers in the test set according to the quartile of the tumor area classified as LumA by the deep learning model (total n = 230)
View article: Supplementary Table S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Table S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
4-part table showing hospital sources for cases in the training, validation, initial testing, and final testing subsets.
View article: Figure 2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Figure 2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Schematic overview of the method for quantifying subtype heterogeneity in held-out whole-slide H&E images from TCGA-BRCA cohort. Conv-layer, convolutional layer; SSGCE, sample-specific generalized cross-entropy.
View article: Figure 1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Figure 1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Diagram depicting the allocation of breast cancer WSIs for analysis. Phase A: filtering of TCGA images. Phase B: 680 cases were divided by assigned subtype, and then subsets with transcriptomically pure subtype adherence were randomly spli…
View article: Supplementary Table S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Table S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Table S2. Clinical and molecular features of PAM50 Luminal A breast cancers in the independent test set (n = 230) according to degree of adherence of transcriptomic profile to the LumA subtype by semi-supervised noon-negative…
View article: Supplementary Figure S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Figure S1 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Figure S1. Scatterplot of LumA proportion by transcriptomic analysis versus percentage of tumor image patches classified as LumA by the DNN model - including held-out cases with nonLumA PAM50 assignment (n = 256). Best-fittin…
View article: Supplementary Figure S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Supplementary Figure S2 from Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Supplementary Figure S2. Paired scatterplots comparing 17 Luminal A cases from the test set obtained from hospitals that did not contribute cases for model training versus the 213 Luminal A cases remaining in the test set. Correlations bet…
View article: FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment
FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment Open
Federated Learning (FL) offers a decentralized paradigm for collaborative model training without direct data sharing, yet it poses unique challenges for Domain Generalization (DG), including strict privacy constraints, non-i.i.d. local dat…
View article: Scalable Whole Slide Image Representation Using K-Mean Clustering and Fisher Vector Aggregation
Scalable Whole Slide Image Representation Using K-Mean Clustering and Fisher Vector Aggregation Open
Whole slide images (WSIs) are high-resolution, gigapixel sized images that pose significant computational challenges for traditional machine learning models due to their size and heterogeneity.In this paper, we present a scalable and effic…
View article: Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images
Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images Open
Intratumor heterogeneity (ITH) presents challenges for precision oncology, but methods for its spatial quantification, scalable at population levels, do not exist. Based on previous work showing that the admixture of PAM50 subtype can be m…
View article: Mitigating Instance-Dependent Label Noise: Integrating Self-Supervised Pretraining with Pseudo-Label Refinement
Mitigating Instance-Dependent Label Noise: Integrating Self-Supervised Pretraining with Pseudo-Label Refinement Open
Deep learning models rely heavily on large volumes of labeled data to achieve high performance. However, real-world datasets often contain noisy labels due to human error, ambiguity, or resource constraints during the annotation process. I…
View article: IDAL: Improved Domain Adaptive Learning for Natural Images Dataset
IDAL: Improved Domain Adaptive Learning for Natural Images Dataset Open
View article: Network Inversion and Its Applications
Network Inversion and Its Applications Open
Neural networks have emerged as powerful tools across various applications, yet their decision-making process often remains opaque, leading to them being perceived as "black boxes." This opacity raises concerns about their interpretability…