Understanding Uncertainty Maps in Vision with Statistical Testing

Nazarovs, Jurijs; Huang, Zhichun; Tasneeyapant, Songwong; Chakraborty, Rudrasis; Singh, Vikas

doi:10.1109/cvpr52688.2022.00050

Cited by 4 publications

(4 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this case, the MSE for interpolation and extrapolation were 0.0082 and 0.1545 respectively. In comparison NODE Nazarovs et al [2021] is 0.0074 and 0.1661 Takeaway. Clearly, we can see that by introducing IGO in the model, we get a good generation ability even for extrapolation in the future time steps.…”

Section: Resultsmentioning

confidence: 88%

“…One of the applications of our proposed Intermediate Generator Optimization is modeling the trajectory of a Differential Equation, which describes the dynamic process. We use the setup in Nazarovs et al [2021], that is, we consider that dynamic process is modeled by changes in the latent space z as ż(t) = D(z, t)w, where D(z, t) is a neural network and w is a corresponding mixed effect (random projection describing flexibility (stochasticity) of dynamics). We propose to extend D(z, t) with our IGO, as D(z t , z τ ) using a neural network f , see Figure 4 -(a) for details on loss computation with z τ .…”

Section: Setup Detailsmentioning

confidence: 99%

“…Applications enabled by DGMs in vision include style transferring Zhu et al [2017], inpainting Demir and Unal [2018], image restoration and manipulation Pan et al [2021]. VAE architectures have been successfully deployed in temporal prediction settings owing to their robustness properties Nazarovs et al [2021], Rubanova et al [2019].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Using Intermediate Forward Iterates for Intermediate Generator Optimization

Mishra¹,

Nazarovs²,

Dogra³

et al. 2023

Preprint

View full text Add to dashboard Cite

Score-based models have recently been introduced as a richer framework to model distributions in high dimensions and are generally more suitable for generative tasks. In score-based models, a generative task is formulated using a parametric model (such as a neural network) to directly learn the gradient of such high dimensional distributions, instead of the density functions themselves, as is done traditionally. From the mathematical point of view, such gradient information can be utilized in reverse by stochastic sampling to generate diverse samples. However, from a computational perspective, existing score-based models can be efficiently trained only if the forward or the corruption process can be computed in closed form. By using the relationship between the process and layers in a feed-forward network, we derive a backpropagation-based procedure which we call Intermediate Generator Optimization to utilize intermediate iterates of the process with negligible computational overhead. The main advantage of IGO is that it can be incorporated into any standard autoencoder pipeline for the generative task. We analyze the sample complexity properties of IGO to solve downstream tasks like Generative PCA. We show applications of the IGO on two dense predictive tasks viz., image extrapolation, and point cloud denoising. Our experiments indicate that obtaining an ensemble of generators for various time points is possible using first-order methods.Preprint. Under review.

show abstract

Section: Resultsmentioning

confidence: 88%

Section: Setup Detailsmentioning

confidence: 99%

See 1 more Smart Citation

Using Intermediate Forward Iterates for Intermediate Generator Optimization

Mishra¹,

Nazarovs²,

Dogra³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Addressing this limitation is planned for future work, and different methods can be used to consider patient specific parameters in a ML framework. [30][31][32] Another limitation, is that due to the lack of publicly available data, we could only use in silico data in this study. Moreover, the performance comparison must be considered with the fact in mind that the PK-SciML framework used the known true structure, which enabled the superior performance.…”

Section: Discussionmentioning

confidence: 99%

Integrating machine learning with pharmacokinetic models: Benefits of scientific machine learning in adding neural networks components to existing PK models

Valderrama,

Ponce‐Bobadilla,

Mensing

et al. 2023

CPT Pharmacom & Syst Pharma

View full text Add to dashboard Cite

Recently, the use of machine‐learning (ML) models for pharmacokinetic (PK) modeling has grown significantly. Although most of the current approaches use ML techniques as black boxes, there are only a few that have proposed interpretable architectures which integrate mechanistic knowledge. In this work, we use as the test case a one‐compartment PK model using a scientific machine learning (SciML) framework and consider learning an unknown absorption using neural networks, while simultaneously estimating other parameters of drug distribution and elimination. We generate simulated data with different sampling strategies to show that our model can accurately predict concentrations in extrapolation tasks, including new dosing regimens with different sparsity levels, and produce reliable forecasts even for new patients. By using a scenario of fitting PK data with complex absorption, we demonstrate that including known physiological structure into an SciML model allows us to obtain highly accurate predictions while preserving the interpretability of classical compartmental models.

show abstract

AdvMIL: Adversarial multiple instance learning for the survival analysis on whole-slide images

Liu,

Ji,

et al. 2024

Medical Image Analysis

View full text Add to dashboard Cite

Understanding Uncertainty Maps in Vision with Statistical Testing

Cited by 4 publications

References 36 publications

Using Intermediate Forward Iterates for Intermediate Generator Optimization

Using Intermediate Forward Iterates for Intermediate Generator Optimization

Integrating machine learning with pharmacokinetic models: Benefits of scientific machine learning in adding neural networks components to existing PK models

AdvMIL: Adversarial multiple instance learning for the survival analysis on whole-slide images

Contact Info

Product

Resources

About