Detecting test samples drawn sufficiently far away from the training distribution statistically or adversarially is a fundamental requirement for deploying good classifier in many real-world machine learning applications. However, deep neural networks with softmax are known to produce highly overconfident posterior distributions even such abnormal samples. In this paper, we propose simple yet effective method detecting any samples, which applicable pre-trained classifier. We obtain class...

The problem of detecting whether a test sample is from in-distribution (i.e., training distribution by classifier) or out-of-distribution sufficiently different it arises in many real-world machine learning applications. However, the state-of-art deep neural networks are known to be highly overconfident their predictions, i.e., do not distinguish in- and out-of-distributions. Recently, handle this issue, several threshold-based detectors have been proposed given pre-trained classifiers....

The popularity of Aloha-like algorithms for resolution contention between multiple entities accessing common resources is due to their extreme simplicity and distributed nature. Example applications such include Ethernet recently emerging wireless multi-access networks. Despite a long exciting history more than four decades, the question designing an algorithm that essentially as simple Aloha while being efficient has remained unresolved.

Deep neural networks with millions of parameters may suffer from poor generalization due to overfitting. To mitigate the issue, we propose a new regularization method that penalizes predictive distribution between similar samples. In particular, distill different samples same label during training. This results in regularizing dark knowledge (i.e., on wrong predictions) single network self-knowledge distillation) by forcing it produce more meaningful and consistent predictions class-wise...

Novelty detection, i.e., identifying whether a given sample is drawn from outside the training distribution, essential for reliable machine learning. To this end, there have been many attempts at learning representation well-suited novelty detection and designing score based on such representation. In paper, we propose simple, yet effective method named contrasting shifted instances (CSI), inspired by recent success contrastive of visual representations. Specifically, in addition to with...

In most real-world scenarios, labeled training datasets are highly class-imbalanced, where deep neural networks suffer from generalizing to a balanced testing criterion. this paper, we explore novel yet simple way alleviate issue by augmenting less-frequent classes via translating samples (e.g., images) more-frequent classes. This approach enables classifier learn more generalizable features of minority classes, transferring and leveraging the diversity majority information. Our experimental...

Lifelong learning with deep neural networks is well-known to suffer from catastrophic forgetting: the performance on previous tasks drastically degrades when a new task. To alleviate this effect, we propose leverage large stream of unlabeled data easily obtainable in wild. In particular, design novel class-incremental scheme (a) distillation loss, termed global distillation, (b) strategy avoid overfitting most recent task, and (c) confidence-based sampling method effectively external data....

Recent breakthroughs in self-supervised learning show that such algorithms learn visual representations can be transferred better to unseen tasks than cross-entropy based methods which rely on task-specific supervision. In this paper, we found the similar holds continual context: contrastively learned are more robust against catastrophic forgetting ones trained with objective. Based novel observation, propose a rehearsal-based algorithm focuses continually and maintaining transferable...

Despite the remarkable progress in deep generative models, synthesizing high-resolution and temporally coherent videos still remains a challenge due to their high-dimensionality complex temporal dynamics along with large spatial variations. Recent works on diffusion models have shown potential solve this challenge, yet they suffer from severe computation memory-inefficiency that limit scalability. To handle issue, we propose novel model for videos, coined projected latent video (PVDM),...

This paper provides proofs of the rate stability, Harris recurrence, and ε-optimality carrier sense multiple access (CSMA) algorithms where random (or backoff) parameter each node is adjusted dynamically. These require only local information they are easy to implement. The setup a network wireless nodes with fixed conflict graph that identifies pairs whose simultaneous transmissions conflict. studies two algorithms. first algorithm schedules keep up given arrival rates packets. second...

There has recently been considerable interest in design of low-complexity, myopic, distributed and stable scheduling algorithms for constrained queueing network models that arise the context emerging communication networks. Here we consider two representative models. One, a model captures randomly varying number packets queues present at collection wireless nodes communicating through shared medium. Two, buffered circuit switched an optical core future internet to capture randomness calls or...

Generative adversarial networks (GANs) have shown outstanding performance on a wide range of problems in computer vision, graphics, and machine learning, but often require numerous training data heavy computational resources. To tackle this issue, several methods introduce transfer learning technique GAN training. They, however, are either prone to overfitting or limited small distribution shifts. In paper, we show that simple fine-tuning GANs with frozen lower layers the discriminator...

Unsupervised image-to-image translation has gained considerable attention due to the recent impressive progress based on generative adversarial networks (GANs). However, previous methods often fail in challenging cases, particular, when an image multiple target instances and a task involves significant changes shape, e.g., translating pants skirts fashion images. To tackle issues, we propose novel method, coined instance-aware GAN (InstaGAN), that incorporates instance information (e.g.,...

Recent improvements in deep learning and hardware support offer a new breakthrough mobile sensing; we could enjoy context-aware services healthcare on device powered by artificial intelligence. However, most related studies perform well only with certain level of similarity between trained target data distribution, while practice, specific user's behaviors make sensor inputs different. Consequently, the performance such applications might suffer diverse user conditions as training models...

While semi-supervised learning (SSL) has proven to be a promising way for leveraging unlabeled data when labeled is scarce, the existing SSL algorithms typically assume that training class distributions are balanced. However, these trained under imbalanced can severely suffer generalizing balanced testing criterion, since they utilize biased pseudo-labels of toward majority classes. To alleviate this issue, we formulate convex optimization problem softly refine generated from model, and...

In the deep learning era, long video generation of high-quality still remains challenging due to spatio-temporal complexity and continuity videos. Existing prior works have attempted model distribution by representing videos as 3D grids RGB values, which impedes scale generated neglects continuous dynamics. this paper, we found that recent emerging paradigm implicit neural representations (INRs) encodes a signal into parameterized network effectively mitigates issue. By utilizing INRs video,...

Internet video streaming has experienced tremendous growth over the last few decades. However, quality of existing delivery critically depends on bandwidth resource. Consequently, user experience (QoE) suffers inevitably when network conditions become unfavorable. We present a new framework that utilizes client computation and recent advances in deep neural networks (DNNs) to reduce dependency for delivering high-quality video. The use DNNs enables us enhance independent available bandwidth....

As the application of deep learning has expanded to real-world problems with insufficient volume training data, transfer recently gained much attention as means improving performance in such small-data regime. However, when existing methods are applied between heterogeneous architectures and tasks, it becomes more important manage their detailed configurations often requires exhaustive tuning on them for desired performance. To address issue, we propose a novel approach based meta-learning...

Deep neural networks have achieved impressive success in large-scale visual object recognition tasks with a predefined set of classes. However, recognizing objects novel classes unseen during training still remains challenging. The problem detecting such has been addressed the literature, but most prior works focused on providing simple binary or regressive decisions, e.g., output would be "known," "novel," corresponding confidence intervals. In this paper, we study more informative novelty...