Conference Papers
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/123456789/28506
Browse
Search Results
Item DROCC: Deep Robust One-Class Classification(ML Research Press, 2020) Goyal, S.; Raghunathan, A.; Jain, M.; Simhadri, H.; Jain, P.Classical approaches for one-class problems such as one-class SVM and isolation forest require careful feature engineering when applied to structured domains like images. State-of-the-art methods aim to leverage deep learning to learn appropriate features via two main approaches. The first approach based on predicting transformations (Golan & El-Yaniv, 2018; Hendrycks et al., 2019a) while successful in some domains, crucially depends on an appropriate domain-specific set of transformations that are hard to obtain in general. The second approach of minimizing a classical one-class loss on the learned final layer representations, e.g., DeepSVDD (Ruff et al., 2018) suffers from the fundamental drawback of representation collapse. In this work, we propose Deep Robust One Class Classification (DROCC) that is both applicable to most standard domains without requiring any side-information and robust to representation collapse. DROCC is based on the assumption that the points from the class of interest lie on a well-sampled, locally linear low dimensional manifold. Empirical evaluation demonstrates that DROCC is highly effective in two different one-class problem settings and on a range of real-world datasets across different domains: tabular data, images (CIFAR and ImageNet), audio, and time-series, offering up to 20% increase in accuracy over the state-of-the-art in anomaly detection. Code is available at https://github.com/microsoft/EdgeML. © 2020 by the author(s).Item DROCC: Deep robust one-class classification(International Machine Learning Society (IMLS), 2020) Goyal, S.; Raghunathan, A.; Jain, M.; Simhadri, H.; Jain, P.Classical approaches for one-class problems such as one-class SVM and isolation forest require careful feature engineering when applied to structured domains like images. State-of-the-art methods aim to leverage deep learning to learn appropriate features via two main approaches. The first approach based on predicting transformations (Golan & El-Yaniv, 2018; Hendrycks et al., 2019a) while successful in some domains, crucially depends on an appropriate domain-specific set of transformations that are hard to obtain in general. The second approach of minimizing a classical one-class loss on the learned final layer representations, e.g., DeepSVDD (Ruff et al., 2018) suffers from the fundamental drawback of representation collapse. In this work, we propose Deep Robust One Class Classification (DROCC) that is both applicable to most standard domains without requiring any side-information and robust to representation collapse. DROCC is based on the assumption that the points from the class of interest lie on a well-sampled, locally linear low dimensional manifold. Empirical evaluation demonstrates that DROCC is highly effective in two different one-class problem settings and on a range of real-world datasets across different domains: tabular data, images (CIFAR and ImageNet), audio, and time-series, offering up to 20% increase in accuracy over the state-of-the-art in anomaly detection. Code is available at https://github.com/microsoft/EdgeML. © 2020 by the author(s).Item Improving convergence in Irgan with PPO(Association for Computing Machinery, 2020) Jain, M.; Kamath S․, S.Information retrieval modeling aims to optimise generative and discriminative retrieval strategies, where, generative retrieval focuses on predicting query-specific relevant documents and discriminative retrieval tries to predict relevancy given a query-document pair. IRGAN unifies the generative and discriminative retrieval approaches through a minimax game. However, training IRGAN is unstable and varies largely with the random initialization of parameters. In this work, we propose improvements to IRGAN training through a novel optimization objective based on proximal policy optimisation and gumbel-softmax based sampling for the generator, along with a modified training algorithm which performs the gradient update on both the models simultaneously for each training iteration. We benchmark our proposed approach against IRGAN on three different information retrieval tasks and present empirical evidence of improved convergence. © 2020 Copyright held by the owner/author(s). Publication rights licensed to ACM.Item Machine Learning Models with Optimization for Clothing Recommendation from Personal Wardrobe(Institute of Electrical and Electronics Engineers Inc., 2020) Jain, M.; Singh, S.; Chandrasekaran, K.; Rathnamma, M.V.; Ramana, V.In the present-day scenario, several clothing recommender systems have been developed for the online e-commerce industry. However, when it comes to recommending clothes that a person already possesses, i.e, from their personal wardrobe, there are very few systems that have been proposed to perform the task. In this paper, we tackle the latter issue, and perform experimental analysis of the various Machine Learning techniques that can be used for carrying out the task. Since the recommendations must be made from a user's personal wardrobe, the recommender system doesn't follow a traditional approach. This is explained in detail in the following sections. Further, the paper contains a complete description of the results obtained from the experiments conducted, and the best approach is specified, with appropriate justification for the same. © 2020 IEEE.