| 1. |
- Fay, Dominik, et al.
(author)
-
Adaptive Hyperparameter Selection for Differentially Private Gradient Descent
- 2023
-
In: Transactions on Machine Learning Research. - 2835-8856.
-
Journal article (peer-reviewed)abstract
- We present an adaptive mechanism for hyperparameter selection in differentially private optimization that addresses the inherent trade-off between utility and privacy. The mechanism eliminates the often unstructured and time-consuming manual effort of selecting hyperparameters and avoids the additional privacy costs that hyperparameter selection otherwise incurs on top of that of the actual algorithm.We instantiate our mechanism for noisy gradient descent on non-convex, convex and strongly convex loss functions, respectively, to derive schedules for the noise variance and step size. These schedules account for the properties of the loss function and adapt to convergence metrics such as the gradient norm. When using these schedules, we show that noisy gradient descent converges at essentially the same rate as its noise-free counterpart. Numerical experiments show that the schedules consistently perform well across a range of datasets without manual tuning.
|
|
| 2. |
- Ancuti, Codruta O., et al.
(author)
-
NTIRE 2023 HR NonHomogeneous Dehazing Challenge Report
- 2023
-
In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). - Vancover : Institute of Electrical and Electronics Engineers (IEEE).
-
Conference paper (peer-reviewed)abstract
- This study assesses the outcomes of the NTIRE 2023 Challenge on Non-Homogeneous Dehazing, wherein novel techniques were proposed and evaluated on new image dataset called HD-NH-HAZE. The HD-NH-HAZE dataset contains 50 high resolution pairs of real-life outdoor images featuring nonhomogeneous hazy images and corresponding haze-free images of the same scene. The nonhomogeneous haze was simulated using a professional setup that replicated real-world conditions of hazy scenarios. The competition had 246 participants and 17 teams that competed in the final testing phase, and the proposed solutions demonstrated the cutting-edge in image dehazing technology.
|
|
| 3. |
- Bånkestad, Maria, et al.
(author)
-
Variational Elliptical Processes
- 2023
-
In: Transactions on Machine Learning Research. - 2835-8856.
-
Journal article (peer-reviewed)abstract
- We present elliptical processes—a family of non-parametric probabilistic models that subsumes Gaussian processes and Student's t processes. This generalization includes a range of new heavy-tailed behaviors while retaining computational tractability. Elliptical processes are based on a representation of elliptical distributions as a continuous mixture of Gaussian distributions. We parameterize this mixture distribution as a spline normalizing flow, which we train using variational inference. The proposed form of the variational posterior enables a sparse variational elliptical process applicable to large-scale problems. We highlight advantages compared to Gaussian processes through regression and classification experiments. Elliptical processes can supersede Gaussian processes in several settings, including cases where the likelihood is non-Gaussian or when accurate tail modeling is essential.
|
|
| 4. |
- Conde, Marcus V., et al.
(author)
-
Lens-to-Lens Bokeh Effect Transformation : NTIRE 2023 Challenge Report
- 2023
-
In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. - Vancover : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 1643-1659
-
Conference paper (other academic/artistic)abstract
- We present the new Bokeh Effect Transformation Dataset (BETD), and review the proposed solutions for this novel task at the NTIRE 2023 Bokeh Effect Transformation Challenge. Recent advancements of mobile photography aim to reach the visual quality of full-frame cameras. Now, a goal in computational photography is to optimize the Bokeh effect itself, which is the aesthetic quality of the blur in out-of-focus areas of an image. Photographers create this aesthetic effect by benefiting from the lens optical properties. The aim of this work is to design a neural network capable of converting the the Bokeh effect of one lens to the effect of another lens without harming the sharp foreground regions in the image. For a given input image, knowing the target lens type, we render or transform the Bokeh effect accordingly to the lens properties. We build the BETD using two full-frame Sony cameras, and diverse lens setups. To the best of our knowledge, we are the first attempt to solve this novel task, and we provide the first BETD dataset and benchmark for it. The challenge had 99 registered participants. The submitted methods gauge the state-of-the-art in Bokeh effect rendering and transformation.
|
|
| 5. |
- Fay, Dominik, et al.
(author)
-
Private Learning Via Knowledge Transfer with High-Dimensional Targets
- 2022
-
In: ICASSP 2022. - : Institute of Electrical and Electronics Engineers (IEEE). - 9781665405409 - 9781665405416 ; , s. 3873-3877
-
Conference paper (peer-reviewed)abstract
- Preventing unintentional leakage of information about the training set has high relevance for many machine learning tasks, such as medical image segmentation. While differential privacy (DP) offers mathematically rigorous protection, the high output dimensionality of segmentation tasks prevents the direct application of state-of-the-art algorithms such as Private Aggregation of Teacher Ensembles (PATE). In order to alleviate this problem, we propose to learn dimensionality-reducing transformations to map the prediction target into a bounded lower-dimensional space to reduce the required noise level during the aggregation stage. To this end, we assess the suitability of principal component analysis (PCA) and autoencoders. We conclude that autoencoders are an effective means to reduce the noise in the target variables.
|
|
| 6. |
- Hering, Alessa, et al.
(author)
-
Learn2Reg: comprehensive multi-task medical image registration challenge, dataset and evaluation in the era of deep learning
- 2023
-
In: IEEE Transactions on Medical Imaging. - : Institute of Electrical and Electronics Engineers (IEEE). - 0278-0062 .- 1558-254X. ; 42:3, s. 697-712
-
Journal article (peer-reviewed)abstract
- Image registration is a fundamental medical image analysis task, and a wide variety of approaches have been proposed. However, only a few studies have comprehensively compared medical image registration approaches on a wide range of clinically relevant tasks. This limits the development of registration methods, the adoption of research advances into practice, and a fair benchmark across competing approaches. The Learn2Reg challenge addresses these limitations by providing a multi-task medical image registration data set for comprehensive characterisation of deformable registration algorithms. A continuous evaluation will be possible at https:// learn2reg.grand-challenge.org. Learn2Reg covers a wide range of anatomies (brain, abdomen, and thorax), modalities (ultrasound, CT, MR), availability of annotations, as well as intra- and inter-patient registration evaluation. We established an easily accessible framework for training and validation of 3D registration methods, which enabled the compilation of results of over 65 individual method submissions from more than 20 unique teams. We used a complementary set of metrics, including robustness, accuracy, plausibility, and runtime, enabling unique insight into the current state-of-the-art of medical image registration. This paper describes datasets, tasks, evaluation methods and results of the challenge, as well as results of further analysis of transferability to new datasets, the importance of label supervision, and resulting bias. While no single approach worked best across all tasks, many methodological aspects could be identified that push the performance of medical image registration to new state-of-the-art performance. Furthermore, we demystified the common belief that conventional registration methods have to be much slower than deep-learning-based methods.
|
|
| 7. |
- Luo, Ziwei, et al.
(author)
-
Image Restoration with Mean-Reverting Stochastic Differential Equations
- 2023
-
In: Proceedings of the 40th International Conference on Machine Learning. ; , s. 23045-23066
-
Conference paper (other academic/artistic)abstract
- This paper presents a stochastic differential equation (SDE) approach for general-purpose image restoration. The key construction consists in a mean-reverting SDE that transforms a high-quality image into a degraded counterpart as a mean state with fixed Gaussian noise. Then, by simulating the corresponding reverse-time SDE, we are able to restore the origin of the low-quality image without relying on any task-specific prior knowledge. Crucially, the proposed mean-reverting SDE has a closed-form solution, allowing us to compute the ground truth time-dependent score and learn it with a neural network. Moreover, we propose a maximum likelihood objective to learn an optimal reverse trajectory that stabilizes the training and improves the restoration results. The experiments show that our proposed method achieves highly competitive performance in quantitative comparisons on image deraining, deblurring, and denoising, setting a new state-of-the-art on two deraining datasets. Finally, the general applicability of our approach is further demonstrated via qualitative results on image super-resolution, inpainting, and dehazing. Code is available at https://github.com/Algolzw/image-restoration-sde.
|
|
| 8. |
- Luo, Ziwei, et al.
(author)
-
Refusion : Enabling Large-Size Realistic Image Restoration With Latent-Space Diffusion Models
- 2023
-
In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). - : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 1680-1691
-
Conference paper (peer-reviewed)abstract
- This work aims to improve the applicability of diffusion models in realistic image restoration. Specifically, we enhance the diffusion model in several aspects such as network architecture, noise level, denoising steps, training image size, and optimizer/scheduler. We show that tuning these hyperparameters allows us to achieve better performance on both distortion and perceptual scores. We also propose a U-Net based latent diffusion model which performs diffusion in a low-resolution latent space while preserving high-resolution information from the original input for the decoding process. Compared to the previous latent-diffusion model which trains a VAE-GAN to compress the image, our proposed U-Net compression strategy is significantly more stable and can recover highly accurate images without relying on adversarial optimization. Importantly, these modifications allow us to apply diffusion models to various image restoration tasks, including real-world shadow removal, HR non-homogeneous dehazing, stereo super-resolution, and bokeh effect transformation. By simply replacing the datasets and slightly changing the noise network, our model, named Refusion, is able to deal with large-size images (e.g., 6000 x 4000 x 3 in HR dehazing) and produces good results on all the above restoration problems. Our Refusion achieves the best perceptual performance in the NTIRE 2023 Image Shadow Removal Challenge and wins 2nd place overall.
|
|
| 9. |
- Szczepankiewicz, Filip, et al.
(author)
-
Cross-term-compensated gradient waveform design for tensor-valued diffusion MRI
- 2021
-
In: Journal of Magnetic Resonance. - : Elsevier BV. - 1090-7807 .- 1096-0856. ; 328
-
Journal article (peer-reviewed)abstract
- Diffusion MRI uses magnetic field gradients to sensitize the signal to the random motion of spins. In addition to the prescribed gradient waveforms, background field gradients contribute to the diffusion weighting and thereby cause an error in the measured signal and consequent parameterization. The most prominent contribution to the error comes from so-called ‘cross-terms.’ In this work we present a novel gradient waveform design that enables diffusion encoding that cancels such cross-terms and yields a more accurate measurement. This is achieved by numerical optimization that maximizes encoding efficiency with a simultaneous constraint on the ‘cross-term sensitivity’ (c = 0). We found that the optimized cross-term-compensated waveforms were superior to previous cross-term-compensated designs for a wide range of waveform types that yield linear, planar, and spherical b-tensor encoding. The efficacy of the proposed design was also demonstrated in practical experiments using a clinical MRI system. The sensitivity to cross-terms was evaluated in a water phantom with a folded surface which provoked strong internal field gradients. In every comparison, the cross-term-compensated waveforms were robust to the effects of background gradients, whereas conventional designs were not. We also propose a method to measure background gradients from diffusion-weighted data, and show that cross-term-compensated waveforms produce parameters that are markedly less dependent on the background compared to non-compensated designs. Finally, we also used simulations to show that the proposed cross-term compensation was robust to background gradients in the interval 0 to 3 mT/m, whereas non-compensated designs were impacted in terms of a severe signal and parameter bias. In conclusion, we have proposed and demonstrated a waveform design that yields efficient cross-term compensation and facilitates accurate diffusion MRI in the presence of static background gradients regardless of their amplitude and direction. The optimization framework is compatible with arbitrary spin-echo sequence timing and RF events, b-tensor shapes, suppression of concomitant gradient effects and motion encoding, and is shared in open source.
|
|
| 10. |
- Szczepankiewicz, Filip, et al.
(author)
-
Motion-compensated gradient waveforms for tensor-valued diffusion encoding by constrained numerical optimization
- 2021
-
In: Magnetic Resonance in Medicine. - : Wiley. - 1522-2594 .- 0740-3194. ; 85:4, s. 2117-2126
-
Journal article (peer-reviewed)abstract
- PURPOSE: Diffusion-weighted MRI is sensitive to incoherent tissue motion, which may confound the measured signal and subsequent analysis. We propose a "motion-compensated" gradient waveform design for tensor-valued diffusion encoding that negates the effects bulk motion and incoherent motion in the ballistic regime.METHODS: Motion compensation was achieved by constraining the magnitude of gradient waveform moment vectors. The constraint was incorporated into a numerical optimization framework, along with existing constraints that account for b-tensor shape, hardware restrictions, and concomitant field gradients. We evaluated the efficacy of encoding and motion compensation in simulations, and we demonstrated the approach by linear and planar b-tensor encoding in a healthy heart in vivo.RESULTS: The optimization framework produced asymmetric motion-compensated waveforms that yielded b-tensors of arbitrary shape with improved efficiency compared with previous designs for tensor-valued encoding, and equivalent efficiency to previous designs for linear (conventional) encoding. Technical feasibility was demonstrated in the heart in vivo, showing vastly improved data quality when using motion compensation. The optimization framework is available online in open source.CONCLUSION: Our gradient waveform design is both more flexible and efficient than previous methods, facilitating tensor-valued diffusion encoding in tissues in which motion would otherwise confound the signal. The proposed design exploits asymmetric encoding times, a single refocusing pulse or multiple refocusing pulses, and integrates compensation for concomitant gradient effects throughout the imaging volume.
|
|
| 11. |
- Vasluianu, Florin -Alexandru, et al.
(author)
-
NTIRE 2023 Image Shadow Removal Challenge Report
- 2023
-
In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). - Vancover : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 1788-1807
-
Conference paper (other academic/artistic)abstract
- This work reviews the results of the NTIRE 2023 Challenge on Image Shadow Removal. The described set of solutions were proposed for a novel dataset, which captures a wide range of object-light interactions. It consists of 1200 roughly pixel aligned pairs of real shadow free and shadow affected images, captured in a controlled environment. The data was captured in a white-box setup, using professional equipment for lights and data acquisition. The challenge had a number of 144 participants registered, out of which 19 teams were compared in the final ranking. The proposed solutions extend the work on shadow removal, improving over well-established state-of-the-art methods.
|
|
| 12. |
- Wang, Longguang, et al.
(author)
-
NTIRE 2023 Challenge on Stereo Image Super-Resolution : Methods and Results
- 2023
-
In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). - Vancover : Institute of Electrical and Electronics Engineers (IEEE). - 9798350302493 - 9798350302509 ; , s. 1346-1372
-
Conference paper (other academic/artistic)abstract
- In this paper, we summarize the 2nd NTIRE challenge on stereo image super-resolution (SR) with a focus on new solutions and results. The task of the challenge is to super-resolve a low-resolution stereo image pair to a high-resolution one with a magnification factor of x4. Compared with single image SR, the major challenge of this challenge lies in how to exploit additional information in another viewpoint and how to maintain stereo consistency in the results. This challenge has 3 tracks, including one track on distortion (e.g., PSNR) and bicubic degradation, one track on perceptual quality (e.g., LPIPS) and bicubic degradation, as well as another track on real degradations. In total, 175, 93, and 103 participants were successfully registered for each track, respectively. In the test phase, 21, 17, and 12 teams successfully submitted results with PSNR (RGB) scores better than the baseline. This challenge establishes a new benchmark for stereo image SR.
|
|
| 13. |
- Yik, Jackie T., et al.
(author)
-
Automated electrolyte formulation and coin cell assembly for high-throughput lithium-ion battery research
- 2023
-
In: Digital Discovery. - : Royal Society of Chemistry. - 2635-098X. ; 2:3, s. 799-808
-
Journal article (peer-reviewed)abstract
- Battery cell assembly and testing in conventional battery research is acknowledged to be heavily time-consuming and often suffers from large cell-to-cell variations. Manual battery cell assembly and electrolyte formulations are prone to introducing errors which confound optimization strategies and upscaling. Herein we present ODACell, an automated electrolyte formulation and battery assembly setup, capable of preparing large batches of coin cells. We demonstrate the feasibility of Li-ion cell assembly in an ambient atmosphere by preparing LiFePO4‖Li4Ti5O12-based full cells with dimethyl sulfoxide-based model electrolyte. Furthermore, the influence of water is investigated to account for the hygroscopic nature of the non-aqueous electrolyte when exposed to ambient atmosphere. The reproducibility tests demonstrate a conservative fail rate of 5%, while the relative standard deviation of the discharge capacity after 10 cycles was 2% for the studied system. The groups with 2 vol% and 4 vol% of added water in the electrolyte showed overlapping performance trends, highlighting the nontrivial relationship between water contaminants in the electrolytes and the cycling performance. Thus, reproducible data are essential to ascertain whether or not there are minor differences in the performance for high-throughput electrolyte screenings. ODACell is broadly applicable to coin cell assembly with liquid electrolytes and therefore presents an essential step towards accelerating research and development of such systems.
|
|
| 14. |
- Zhao, Zheng, et al.
(author)
-
Probabilistic Estimation of Instantaneous Frequencies of Chirp Signals
- 2023
-
In: IEEE Transactions on Signal Processing. - : Institute of Electrical and Electronics Engineers (IEEE). - 1053-587X .- 1941-0476. ; 71, s. 461-476
-
Journal article (peer-reviewed)abstract
- We present a continuous-time probabilistic approach for estimating the chirp signal and its instantaneous frequency function when the true forms of these functions are not accessible. Our model represents these functions by non-linearly cascaded Gaussian processes represented as non-linear stochastic differential equations. The posterior distribution of the functions is then estimated with stochastic filters and smoothers. We compute a (posterior) Cramér-Rao lower bound for the Gaussian process model, and derive a theoretical upper bound for the estimation error in the mean squared sense. The experiments show that the proposed method outperforms a number of state-of-the-art methods on a synthetic data. We also show that the method works out-of-the-box for two real-world datasets.
|
|
