Abstract

Invertible convolutions have been an essential element for building expressive normalizing flow-based generative models since their introduction in Glow. Several attempts have been made to design invertible k × k convolutions that are efficient in training and sampling passes. Though these attempts have improved the expressivity and sampling efficiency, they severely lagged behind Glow which used only 1×1 convolutions in terms of sampling time. Also, many of the approaches mask a large number of parameters of the underlying convolution, resulting in lower expressivity on a fixed run-time budget. We propose a k × k convolutional layer and Deep Normalizing Flow architecture which i.) has a fast parallel inversion algorithm with running time O(nk2 ) (n is height and width of the input image and k is kernel size), ii.) masks the minimal amount of learnable parameters in a layer. iii.) gives better forward pass and sampling times comparable to other k ×k convolution-based models on real-world benchmarks. We provide an implementation of the proposed parallel algorithm for sampling using our invertible convolutions on GPUs. Benchmarks on CIFAR-10, ImageNet, and CelebA datasets show comparable performance to previous works regarding bits per dimension while significantly improving the sampling time.

Abstract

Sparse neural networks have been proven to generate efficient and better runtimes when compared to dense neural networks. Accelera- tion in runtime is better achieved with structured sparsity. However, generating an efficient sparsity structure to maintain both runtime and accuracy is a challenging task. In this paper, we implement the RBGP4 sparsity pattern derived from the Ramanujan Bipartite Graph Product (RBGP) framework on various Computer Vision tasks and test how well it performs w.r.t accuracy and runtime. Us- ing this approach, we generate structured sparse neural networks which has multiple levels of block sparsity that generates good connectivity due to the presence of Ramanujan bipartite graphs. We benchmark our approach on Semantic Segmentation and Pose Estimation tasks on an edge device (Jetson Nano 2GB) as well as server (V100) GPUs. We compare the results obtained for RBGP4 sparsity pattern with the unstructured and block sparsity patterns. When compared to sparsity patterns like unstructured and block, we obtained significant speedups while maintaining accuracy. KEYWORDS Sparse Neural Networks, Ramanujan Graphs, Semantic Segmenta- tion, Pose Detection ACM Reference Format: Thejasvi Konduru, Furqan Ahmed Shaik, Girish Varma, and Kishore Kotha- palli. 2023. Accelerating Computer Vision Tasks on GPUs using Ramanujan Graph Product Framework. In 6th Joint International Conference on Data Science & Management of Data (10th ACM IKDD CODS and 28th COMAD) (CODS-COMAD 2023), January 4–7, 2023, Mumbai, India. ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3570991.3571044

Abstract

Melanin is a major contributor to the intensity and width of the retinal pigment epithelium (RPE) on optical coherence tomography (OCT). It is hypothesized, high melanin concentrations cause scattering making the RPE appear as a thick band and obscures adjacent layers. A split band appearance of the outermost hyperreflective band has been demonstrated in Albinism, meaning the RPE and Bruch’s Membrane appear as two separate bands. We performed a cross-sectional observational study to test the hypothesis that a reduction in melanin causes an increased ability to differentiate the hyperreflective bands seen on OCT images of the outer retina.

Abstract

urpose: Melanin is a major contributor to the intensity and width of the retinal pigment epithelium (RPE) on optical coherence tomography (OCT). It is hypothesized, high melanin concentrations cause scattering making the RPE appear as a thick band and obscures adjacent layers. A split band appearance of the outermost hyperreflective band has been demonstrated in Albinism, meaning the RPE and Bruch’s Membrane appear as two separate bands. We performed a cross-sectional observational study to test the hypothesis that a reduction in melanin causes an increased ability to differentiate the hyperreflective bands seen on OCT images of the outer retina. Methods: Using the UK Biobank dataset, we generated a randomised sample of 300 participants with specific inclusion and exclusion criteria. Inclusion criteria was an age between 40 to 65, visual acuity better than 0.60 logMAR, no ocular pathology, and

Abstract

Normalizing flows are an essential alternative to GANs for generative modelling, which can be optimized directly on the maximum likelihood of the dataset. They also allow computation of the exact latent vector corresponding to an image since they are composed of invertible transformations. However, the requirement of invertibility of the transformation prevents standard and expressive neural network models such as CNNs from being directly used. Emergent convolutions were proposed to construct an invertible 3×3 CNN layer using a pair of masked CNN layers, making them inefficient. We study conditions such that 3×3 CNNs are invertible, allowing them to construct expressive normalizing flows. We derive necessary and sufficient conditions on a padded CNN for it to be invertible. Our conditions for invertibility are simple, can easily be maintained during the training process. Since we require only a single CNN layer for every effective invertible CNN layer, our approach is more efficient than emerging convolutions. We also proposed a coupling method, Quad-coupling. We benchmark our approach and show similar performance results to emergent convolutions while improving the model's efficiency

Abstract

Quality assessment of agricultural produce is a crucial step in minimizing food stock wastage. However, this is currently done manually and often requires expert supervision, especially in smaller seeds like corn. We propose a novel computer vision-based system for automating this process. We build a novel seed image acquisition setup, which captures both the top and bottom views. Dataset collection for this problem has challenges of data annotation costs/time and class imbalance. We address these challenges by i.) using a Conditional Generative Adversarial Network (CGAN) to generate real-looking images for the classes with lesser images and ii.) annotate a large dataset with minimal expert human intervention by using a Batch Active Learning (BAL) based annotation tool. We benchmark different image classification models on the dataset obtained. We are able to get accuracies of up to 91.6% for testing the physical purity of seed samples.

Abstract

Parametric path problems arise independently in diverse domains, ranging from transportation to finance, where they are studied under various assumptions. We formulate a general path problem with relaxed assumptions, and describe how this formulation is applicable in these domains. We study the complexity of the general problem, and a variant of it where preprocessing is allowed. We show that when the parametric weights are linear functions, algorithms remain tractable even under our relaxed assumptions. Furthermore, we show that if the weights are allowed to be non-linear, the problem becomes NP-hard. We also study the mutli-dimensional version of the problem where the weight functions are parameterized by multiple parameters. We show that even with two parameters, the problem is NP-hard.

Abstract

We continue the study of the covering complexity of constraint satisfaction problems (CSPs) initiated by Guruswami, Håstad and Sudan [SIAM J. Comp. 2002] and Dinur and Kol [CCC’13]. The covering number of a CSP instance Φ is the smallest number of assignments to the variables of Φ, such that each constraint of Φ is satisfied by at least one of the assignments. We show the following results: 1. Assuming a covering variant of the Unique Games Conjecture, introduced by Dinur and Kol, we show that for every non-odd predicate P over any constant-size alphabet and every integer K, it is NP-hard to approximate the covering number within a factor of K. This yields a complete characterization of CSPs over constant-size alphabets that are hard to cover.

Abstract

Recent years have witnessed utilization of modern machine learning approaches for predicting properties of material using available datasets. However, to identify potential candidates for material discovery, one has to systematically scan through a large chemical space and subsequently calculate the properties of all such samples. On the other hand, generative methods are capable of efficiently sampling the chemical space and can generate molecules/materials with desired properties. In this study, we report a deep learning based inorganic material generator (DING) framework consisting of a generator module and a predictor module. The generator module is developed based upon conditional variational autoencoders (CVAE) and the predictor module consists of three deep neural networks trained for predicting enthalpy of formation, volume per atom and energy per atom chosen to demonstrate the proposed method. The predictor and generator modules have been developed using a one hot key representation of the material composition. A series of tests were done to examine the robustness of the predictor models, to demonstrate the continuity of the latent material space, and its ability to generate materials exhibiting target property values. The DING architecture proposed in this paper can be extended to other properties based on which the chemical space can be efficiently explored for interesting materials/molecules.

Abstract

Sparse neural networks are efficient in both memory and compute when compared to dense neural networks. But on parallel hardware such as GPU, sparse neural networks result in small or no runtime performance gains. On the other hand, structured sparsity patterns like filter, channel and block sparsity result in large performance gains due to regularity induced by structure. Among structured sparsities, block sparsity is a generic structured sparsity pattern with filter and channel sparsity being sub cases of block sparsity. In this work, we focus on block sparsity and generate efficient block sparse convolutional neural networks using our approach DBSR (Dynamic block sparse reparameterization). Our DBSR approach, when applied on image classification task over Imagenet dataset, decreases parameters and FLOPS of ResneXt50 by a factor of 2x with only increase of 0.48 in Top-1 error. And when extended to the task of semantic segmentation, our approach reduces parameters and FLOPS by 30% and 20% respectively with only 1% decrease in mIoU for ERFNet over Cityscapes dataset