New adversarial attack method based on deep feature distributions.
problem Adversarial attacks on CNN classifiers using output layer information.
method Modeling and exploiting class-wise and layer-wise deep feature distributions.
result Achieves state-of-the-art transfer-based attack results for undefended ImageNet models.
We present a principled approach for detecting out-of-distribution (OOD) and adversarial samples in deep neural networks. Our approach consists in modeling the outputs of the various layers (deep features) with parametric probability distributions once training is completed. At inference, the likelihoods of the deep fe…
PFDL improves deep learning models' OOD generalization by decorrelating feature embeddings.
problem Out-of-distribution generalization in deep learning models.
method PFDL algorithm that optimizes feature decomposition network and image classification model.
result PFDL improves the accuracy of image classification models on OOD datasets.
The use of distributions and high-level features from deep architecture has become commonplace in modern computer vision. Both of these methodologies have separately achieved a great deal of success in many computer vision tasks. However, there has been little work attempting to leverage the power of these to methodolo…
Herein, we generate pseudo-features based on the multivariate probability distributions obtained from the feature maps in layers of trained deep neural networks. Further, we augment the minor-class data based on these generated pseudo-features to overcome the imbalanced data problems. The proposed method, i.e., cavity …
We propose a neural hybrid model consisting of a linear model defined on a set of features computed by a deep, invertible transformation (i.e. a normalizing flow). An attractive property of our model is that both p(features), the density of the features, and p(targets | features), the predictive distribution, can be co…
Deep metric learning employs deep neural networks to embed instances into a metric space such that distances between instances of the same class are small and distances between instances from different classes are large. In most existing deep metric learning techniques, the embedding of an instance is given by a featur…
In recent work, it was shown that combining multi-kernel based support vector machines (SVMs) can lead to near state-of-the-art performance on an action recognition dataset (HMDB-51 dataset). This was 0.4\% lower than frameworks that used hand-crafted features in addition to the deep convolutional feature extractors. I…
Improves deep learning training by matching mini-batch distributions.
problem Overfitting and noise in mini-batch training.
method ITDM, which matches the moments of mini-batch distributions to reduce overfitting.
result ITDM reduces overfitting and improves DNN training.
Dida learns meta-features invariant to feature permutations.
problem Learning invariant meta-features from dataset samples.
method Distribution-based deep architecture inheriting universal approximation and robustness.
result Dida outperforms state-of-the-art methods on dataset-level tasks.
New methods improve anomaly detection in deep networks by leveraging hierarchical likelihoods and multi-scale features.
problem Challenges in detecting anomalies in high-level features due to model bias and domain prior.
method Two methods: 1) Log likelihood ratios between in-distribution and general distribution models, 2) Multi-scale likelihood contribution.
result Strong anomaly detection performance in unsupervised setting, slightly underperforming supervised methods.
Hierarchical VAEs detect out-of-distribution data by identifying low-level in-distribution features.
problem Out-of-distribution data often has in-distribution low-level features, leading to misleading likelihood estimates in deep generative models.
method Developed a fast, scalable, unsupervised likelihood-ratio score for out-of-distribution detection based on hierarchical variational autoencoders.
result Achieved state-of-the-art results on out-of-distribution detection across various data and model combinations.
Deep kernel processes unify various models using Gram matrices and kernel functions.
problem Unified representation of various deep learning models.
method Defining deep kernel processes with progressively transformed Gram matrices and sampling from inverse Wishart distributions.
result Deep Gaussian processes, BNNs, infinite BNNs, and infinite BNNs with bottlenecks can all be written as deep kernel processes.
Recently, considerable effort has been devoted to deep domain adaptation in computer vision and machine learning communities. However, most of existing work only concentrates on learning shared feature representation by minimizing the distribution discrepancy across different domains. Due to the fact that all the domai…
Deep models predict intraday electricity prices accurately.
problem Accurately forecasting intraday electricity prices.
method Two deep time series probabilistic models using ESNs with stochastic disturbances and copulas.
result Deep distributional models provide accurate short-term probabilistic price forecasts.
Privacy-preserving distributed deep learning method for multiple classification.
problem Privacy issues in training deep learning models for various fields.
method Split learning into common extractor, cloud model, and local classifier.
result Average performance improvement of 2.63% over existing local training models.
SGD in DLNs reveals feature learning dynamics.
problem Understanding SGD dynamics in DLNs during saddle-to-saddle training.
method Stochastic Langevin dynamics with anisotropic, state-dependent noise; one-dimensional per-mode SDEs; Boltzmann distribution approximation.
result SGD noise encodes feature learning progression but does not alter saddle-to-saddle dynamics.
Deep nearest neighbors outperform self-supervised methods in anomaly detection.
problem Anomaly detection using self-supervised deep methods.
method Simple nearest-neighbor approach on Imagenet pretrained features.
result Nearest-neighbor method outperforms self-supervised methods in accuracy, few shot generalization, training time, and noise robustness.
DeepBDC improves few-shot classification by measuring joint distributions of image features.
problem Few-shot classification with limited training data.
method DeepBDC method using deep learning and Brownian Distance Covariance.
result DeepBDC significantly outperforms existing methods on various benchmarks.
New tool analyzes how features interact with data and models.
problem Lack of theoretical understanding of features in deep learning.
method Introduced interaction tensor to empirically analyze feature interactions.
result Proposed conceptual framework for feature learning with closed-form derivations.
Statistical field theory aids in understanding deep learning complexities.
problem Complexity and lack of theoretical understanding in deep learning.
method Statistical field theory as a theoretical framework.
result Field theory provides insights into generalization, bias, and feature learning.
FeAT improves OOD generalization by learning richer features.
problem Improving feature learning for out-of-distribution (OOD) generalization.
method Feature Augmented Training (FeAT) iteratively augments and retains features from different subsets of training data.
result FeAT effectively learns richer features, boosting OOD performance.
Study on Bayesian deep linear networks with multiple outputs and convolutional layers.
problem Characterize feature learning in finite-width Bayesian deep linear networks.
method Exact and analytical formulas for joint and posterior distributions, using large deviation theory.
result Quantitative description of feature learning in infinite-width regime.
DeepFS uses deep neural networks to select significant features in ultra high-dimensional data.
problem Challenges in traditional feature selection methods for high-dimensional, low-sample-size data.
method Two-step nonparametric approach combining deep neural networks and feature screening.
result DeepFS effectively identifies significant features with high precision for ultra high-dimensional data.
Deep networks, composed of multiple layers of hierarchical distributed representations, tend to learn low-level features in initial layers and transition to high-level features towards final layers. Paradigms such as transfer learning, multi-task learning, and continual learning leverage this notion of generic hierarch…
MixMOOD improves SSDL by selecting unlabelled data based on deep feature similarity.
problem Class distribution mismatch in semi-supervised learning.
method MixMOOD uses deep dataset dissimilarity measures to select unlabelled data.
result MixMOOD selects unlabelled data based on strong correlation with MixMatch accuracy.
Bayesian framework improves deep classifier reliability.
problem Overconfident models under dataset shift.
method Bayesian inference with out-of-distribution data augmentation.
result Reliable uncertainty estimates for deep classifiers.
We propose a novel adversarial speaker adaptation (ASA) scheme, in which adversarial learning is applied to regularize the distribution of deep hidden features in a speaker-dependent (SD) deep neural network (DNN) acoustic model to be close to that of a fixed speaker-independent (SI) DNN acoustic model during adaptatio…
This paper examines how optimization methods affect the reliability of detecting inputs outside a model's training distribution.
problem The unreliability of deep neural networks on out-of-distribution inputs.
method Analysis of optimization methods' impact on OOD detection approaches.
result Optimization methods significantly influence the robustness of OOD detection approaches.
The paper examines the potential of deep learning to support decisions in financial risk management. We develop a deep learning model for predicting whether individual spread traders secure profits from future trades. This task embodies typical modeling challenges faced in risk and behavior forecasting. Conventional ma…
A new method for deep learning under distribution shift by iteratively refining importance weighting.
problem Handling distribution shift in deep learning models when training and test data distributions differ.
method Dynamic Importance Weighting (dynamic IW) that iterates between weight estimation and weighted classification, using a pre-trained feature extractor and stochastic optimization.
result Dynamic IW outperforms state-of-the-art methods in experiments with various types of distribution shift on multiple datasets.
Understanding how species are distributed across landscapes over time is a fundamental question in biodiversity research. Unfortunately, most species distribution models only target a single species at a time, despite strong ecological evidence that species are not independently distributed. We propose Deep Multi-Speci…
Enhances model OOD detection by diversifying feature representation.
problem Limited diversity in model ensembles for OOD detection.
method Integrates diverse training tasks into model ensembles to generate distinct feature representations.
result Superior performance of Multi-Comprehension Ensemble in OOD detection.
New approach uses deep generative models for inventory and pricing decisions.
problem Data-driven inventory and pricing decisions in feature-based newsvendor problems.
method Conditional deep generative models (cDGMs) to learn demand distribution and generate probabilistic forecasts.
result Effective in optimizing inventory and pricing decisions, with theoretical guarantees and real-world applications.
Work proposes a new framework to improve uncertainty estimation in deep Bayesian models.
problem Traditional training procedures underestimate uncertainty in NLMs, leading to unreliable predictions.
method Introduces a novel training framework that captures useful predictive uncertainties for out-of-distribution inputs.
result Demonstrates that traditional methods for NLMs significantly underestimate uncertainty and propose a new framework to address this issue.
A new method generates natural-looking adversarial examples by bounding internal activation values.
problem Creating natural-looking adversarial examples that closely mimic the original input.
method Bounding internal activation values through a distribution quantile bound and polynomial barrier loss function.
result Our attack achieves similar success and confidence levels as state-of-the-art methods but with more natural-looking perturbations.
Paper proposes a GPU-based system for training massive deep learning models in ads systems.
problem Training massive deep learning models with terabyte-scale parameters in ads systems.
method Hierarchical GPU parameter server with 3-layer storage (GPU High-Bandwidth Memory, CPU main memory, SSD).
result 4-node hierarchical GPU parameter server trains a model 2X faster than a 150-node in-memory system.
DFIV uses deep features for IV regression, achieving optimal rates.
problem Optimal IV regression with deep features for complex target functions.
method Two-stage approach: deep feature learning followed by IV regression.
result DFIV achieves minimax optimal learning rate under certain conditions.
ARTree uses deep learning to infer tree topologies efficiently.
problem Efficient phylogenetic inference from tree topologies.
method Deep autoregressive model based on graph neural networks (GNNs).
result ARTree provides a flexible family of distributions over tree topologies.
Study reveals how neural network architectures bias their learning based on feature directions.
problem Understanding how neural network architectures bias their learning based on feature directions.
method Defined neural anisotropy directions (NADs) to encapsulate the directional inductive bias of architectures and provided an efficient method to identify them.
result NADs characterize the features used by CNNs to discriminate between different classes for the CIFAR-10 dataset.
New insights on how weight structure affects generalization in deep Gaussian feature models.
problem Understanding how weight structure impacts generalization in deep learning models.
method Using the replica trick from statistical physics to derive learning curves for models with structured Gaussian features.
result Allowing correlations between the rows of the first layer of features can aid generalization, while structure in later layers is generally detrimental.
Study on shortcuts in deep networks, revealing their layer-wise distribution and impact.
problem Understudied impact of shortcuts on feature representations in deep networks.
method Layer-wise localization through counterfactual training on clean and skewed datasets.
result Shortcuts are distributed throughout the network, not localized in specific layers.
New method calibrates deep models for both in-distribution and out-of-distribution samples.
problem Ensuring calibration for deep models in safety-critical applications, especially in OOD regions.
method Geodesic distance and Gaussian kernel to calibrate deep models.
result Proposed KDF and KDN methods achieve well-calibrated posteriors for both in-distribution and out-of-distribution samples.
Optical Coherence Tomography allows ophthalmologist to obtain cross-section imaging of eye retina. Assisted with digital image analysis methods, effective disease detection could be performed. Various methods exist to extract feature from OCT images. The proposed study aims to compare the effectiveness of handcrafted a…
Deep Gaussian Processes are reinterpreted as deep trigonometric networks for tractable inference.
problem Challenging inference in DGPs due to intractable marginalization in latent function space.
method Viewing DGPs as deep trigonometric networks with Bochner's theorem, and using the wide limit with a bottleneck to translate DGPs into deep trigonometric networks.
result The weight space view yields the same effective covariance functions as obtained in function space, and varying prior distributions over network parameters is equivalent to employing different kernels.
Study analyzes feedback complexity for sparse feature retrieval in deep networks.
problem Learning sparse superposed features with feedback.
method Analysis of feedback complexity in sparse settings, including triplet comparisons.
result Establishes tight bounds and strong upper bounds for feature recovery.
In many important machine learning applications, the training distribution used to learn a probabilistic classifier differs from the testing distribution on which the classifier will be used to make predictions. Traditional methods correct the distribution shift by reweighting the training data with the ratio of the de…
Unsupervised domain adaptation methods aim to alleviate performance degradation caused by domain-shift by learning domain-invariant representations. Existing deep domain adaptation methods focus on holistic feature alignment by matching source and target holistic feature distributions, without considering local feature…