To reach high performance with deep learning, hyperparameter optimization (HPO) is essential. This process is usually time-consuming due to costly evaluations of neural networks. Early discarding techniques limit the resources granted to unpromising candidates by observing the empirical learning curves and canceling neural network training as soon as the lack of competitiveness of a candidate becomes evident. Despite two decades of research, little is understood about the trade-off between the aggressiveness of discarding and the loss of predictive performance. Our paper studies this trade-off for several commonly used discarding techniques such as successive halving and learning curve extrapolation. Our surprising finding is that these commonly used techniques offer minimal to no added value compared to the simple strategy of discarding after a constant number of epochs of training. The chosen number of epochs depends mostly on the available compute budget. We call this approach i-Epoch (i being the constant number of epochs with which neural networks are trained) and suggest to assess the quality of early discarding techniques by comparing how their Pareto-Front (in consumed training epochs and predictive performance) complement the Pareto-Front of i-Epoch.
相關內容
Triangle counting and sampling are two fundamental problems for streaming algorithms. Arguably, designing sampling algorithms is more challenging than their counting variants. It may be noted that triangle counting has received far greater attention in the literature than the sampling variant. In this work, we consider the problem of approximately sampling triangles in different models of streaming with the focus being on the adjacency list model. In this problem, the edges of a graph $G$ will arrive over a data stream. The goal is to design efficient streaming algorithms that can sample and output a triangle from a distribution, over the triangles in $G$, that is close to the uniform distribution over the triangles in $G$. The distance between distributions is measured in terms of $\ell_1$-distance. The main technical contribution of this paper is to design algorithms for this triangle sampling problem in the adjacency list model with the space complexities matching their counting variants. For the sake of completeness, we also show results on the vertex and edge arrival models.
Anomaly detection and localization without any manual annotations and prior knowledge is a challenging task under the setting of unsupervised learning. The existing works achieve excellent performance in the anomaly detection, but with complex networks or cumbersome pipelines. To address this issue, this paper explores a simple but effective architecture in the anomaly detection. It consists of a well pre-trained encoder to extract hierarchical feature representations and a decoder to reconstruct these intermediate features from the encoder. In particular, it does not require any data augmentations and anomalous images for training. The anomalies can be detected when the decoder fails to reconstruct features well, and then errors of hierarchical feature reconstruction are aggregated into an anomaly map to achieve anomaly localization. The difference comparison between those features of encoder and decode lead to more accurate and robust localization results than the comparison in single feature or pixel-by-pixel comparison in the conventional works. Experiment results show that the proposed method outperforms the state-of-the-art methods on MNIST, Fashion-MNIST, CIFAR-10, and MVTec Anomaly Detection datasets on both anomaly detection and localization.
Speaker Embeddings With Weakly Supervised Voice Activity Detection For Efficient Speaker Diarization
Current speaker diarization systems rely on an external voice activity detection model prior to speaker embedding extraction on the detected speech segments. In this paper, we establish that the attention system of a speaker embedding extractor acts as a weakly supervised internal VAD model and performs equally or better than comparable supervised VAD systems. Subsequently, speaker diarization can be performed efficiently by extracting the VAD logits and corresponding speaker embedding simultaneously, alleviating the need and computational overhead of an external VAD model. We provide an extensive analysis of the behavior of the frame-level attention system in current speaker verification models and propose a novel speaker diarization pipeline using ECAPA2 speaker embeddings for both VAD and embedding extraction. The proposed strategy gains state-of-the-art performance on the AMI, VoxConverse and DIHARD III diarization benchmarks.
We are interested in the problem of learning the directed acyclic graph (DAG) when data are generated from a linear structural equation model (SEM) and the causal structure can be characterized by a polytree. Under the Gaussian polytree models, we study sufficient conditions on the sample sizes for the well-known Chow-Liu algorithm to exactly recover both the skeleton and the equivalence class of the polytree, which is uniquely represented by a CPDAG. On the other hand, necessary conditions on the required sample sizes for both skeleton and CPDAG recovery are also derived in terms of information-theoretic lower bounds, which match the respective sufficient conditions and thereby give a sharp characterization of the difficulty of these tasks. We also consider the problem of inverse correlation matrix estimation under the linear polytree models, and establish the estimation error bound in terms of the dimension and the total number of v-structures. We also consider an extension of group linear polytree models, in which each node represents a group of variables. Our theoretical findings are illustrated by comprehensive numerical simulations, and experiments on benchmark data also demonstrate the robustness of polytree learning when the true graphical structures can only be approximated by polytrees.
This work establishes a novel link between the problem of PAC-learning high-dimensional graphical models and the task of (efficient) counting and sampling of graph structures, using an online learning framework. We observe that if we apply the exponentially weighted average (EWA) or randomized weighted majority (RWM) forecasters on a sequence of samples from a distribution P using the log loss function, the average regret incurred by the forecaster's predictions can be used to bound the expected KL divergence between P and the predictions. Known regret bounds for EWA and RWM then yield new sample complexity bounds for learning Bayes nets. Moreover, these algorithms can be made computationally efficient for several interesting classes of Bayes nets. Specifically, we give a new sample-optimal and polynomial time learning algorithm with respect to trees of unknown structure and the first polynomial sample and time algorithm for learning with respect to Bayes nets over a given chordal skeleton.
Traditionally, classical numerical schemes have been employed to solve partial differential equations (PDEs) using computational methods. Recently, neural network-based methods have emerged. Despite these advancements, neural network-based methods, such as physics-informed neural networks (PINNs) and neural operators, exhibit deficiencies in robustness and generalization. To address these issues, numerous studies have integrated classical numerical frameworks with machine learning techniques, incorporating neural networks into parts of traditional numerical methods. In this study, we focus on hyperbolic conservation laws by replacing traditional numerical fluxes with neural operators. To this end, we developed loss functions inspired by established numerical schemes related to conservation laws and approximated numerical fluxes using Fourier neural operators (FNOs). Our experiments demonstrated that our approach combines the strengths of both traditional numerical schemes and FNOs, outperforming standard FNO methods in several respects. For instance, we demonstrate that our method is robust, has resolution invariance, and is feasible as a data-driven method. In particular, our method can make continuous predictions over time and exhibits superior generalization capabilities with out-of-distribution (OOD) samples, which are challenges that existing neural operator methods encounter.
Replica exchange stochastic gradient Langevin dynamics (reSGLD) is an effective sampler for non-convex learning in large-scale datasets. However, the simulation may encounter stagnation issues when the high-temperature chain delves too deeply into the distribution tails. To tackle this issue, we propose reflected reSGLD (r2SGLD): an algorithm tailored for constrained non-convex exploration by utilizing reflection steps within a bounded domain. Theoretically, we observe that reducing the diameter of the domain enhances mixing rates, exhibiting a \emph{quadratic} behavior. Empirically, we test its performance through extensive experiments, including identifying dynamical systems with physical constraints, simulations of constrained multi-modal distributions, and image classification tasks. The theoretical and empirical findings highlight the crucial role of constrained exploration in improving the simulation efficiency.
Understanding degraded speech is demanding, requiring increased listening effort (LE). Evaluating processed and unprocessed speech with respect to LE can objectively indicate if speech enhancement systems benefit listeners. However, existing methods for measuring LE are complex and not widely applicable. In this study, we propose a simple method to evaluate speech intelligibility and LE simultaneously without additional strain on subjects or operators. We assess this method using results from two independent studies in Norway and Denmark, testing 76 (50+26) subjects across 9 (6+3) processing conditions. Despite differences in evaluation setups, subject recruitment, and processing systems, trends are strikingly similar, demonstrating the proposed method's robustness and ease of implementation into existing practices.
This study explores the application of recurrent neural networks to recognize emotions conveyed in music, aiming to enhance music recommendation systems and support therapeutic interventions by tailoring music to fit listeners' emotional states. We utilize Russell's Emotion Quadrant to categorize music into four distinct emotional regions and develop models capable of accurately predicting these categories. Our approach involves extracting a comprehensive set of audio features using Librosa and applying various recurrent neural network architectures, including standard RNNs, Bidirectional RNNs, and Long Short-Term Memory (LSTM) networks. Initial experiments are conducted using a dataset of 900 audio clips, labeled according to the emotional quadrants. We compare the performance of our neural network models against a set of baseline classifiers and analyze their effectiveness in capturing the temporal dynamics inherent in musical expression. The results indicate that simpler RNN architectures may perform comparably or even superiorly to more complex models, particularly in smaller datasets. We've also applied the following experiments on larger datasets: one is augmented based on our original dataset, and the other is from other sources. This research not only enhances our understanding of the emotional impact of music but also demonstrates the potential of neural networks in creating more personalized and emotionally resonant music recommendation and therapy systems.
Graph Neural Networks (GNN) is an emerging field for learning on non-Euclidean data. Recently, there has been increased interest in designing GNN that scales to large graphs. Most existing methods use "graph sampling" or "layer-wise sampling" techniques to reduce training time. However, these methods still suffer from degrading performance and scalability problems when applying to graphs with billions of edges. This paper presents GBP, a scalable GNN that utilizes a localized bidirectional propagation process from both the feature vectors and the training/testing nodes. Theoretical analysis shows that GBP is the first method that achieves sub-linear time complexity for both the precomputation and the training phases. An extensive empirical study demonstrates that GBP achieves state-of-the-art performance with significantly less training/testing time. Most notably, GBP can deliver superior performance on a graph with over 60 million nodes and 1.8 billion edges in less than half an hour on a single machine.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191