Taylorized training improves neural network training at finite width.
problem Understanding and improving neural network training at finite width.
method Training the k-th order Taylor expansion of the neural network at initialization.
result Taylorized training agrees with full neural network training better as k increases and can significantly close the performance gap.
Real-time pruning during training reduces network size and training time.
problem Efficiently reducing neural network size without sacrificing accuracy.
method Activation density-based pruning during training.
result Up to 200x reduction in parameters and 60x reduction in inference compute operations.
ReLU neural networks define piecewise linear functions of their inputs. However, initializing and training a neural network is very different from fitting a linear spline. In this paper, we expand empirically upon previous theoretical work to demonstrate features of trained neural networks. Standard network initializat…
Adaptive networks improve model robustness through conditional normalization.
problem Limited robustness of adversarial-trained networks due to network capacity and training samples.
method Proposes a conditional normalization module to adapt networks during adversarial training.
result Adaptive networks outperform both clean validation accuracy and robustness compared to non-adaptive counterparts.
Dynamic Sparse Training finds efficient sparse networks from scratch.
problem Finding efficient sparse neural networks.
method Jointly optimizes network parameters and sparsity with trainable thresholds.
result Achieves state-of-the-art performance with minimal performance loss.
Un-trained neural networks outperform trained methods in MRI reconstruction.
problem Accelerated MRI reconstruction with minimal training data.
method Variation of Deep Decoder without training data.
result Un-trained approach significantly outperforms other methods in reconstruction accuracy.
Modified RV-coefficient reveals how training affects neural network representations.
problem Understanding how training affects intermediate representations in convolutional neural networks.
method Experimented with modified RV-coefficient (RV2) to compare activation patterns in deep networks trained on varying amounts of data and layers.
result RV2 successfully recovered expected similarity patterns and provided interpretable similarity matrices.
New method reconstructs significant parts of training data from neural networks.
problem Understanding and reconstructing training data from neural networks.
method Proposes a novel reconstruction scheme based on recent theoretical results about neural network training.
result Shows that a significant fraction of training data can be reconstructed from neural network parameters.
It is well-known that neural networks are computationally hard to train. On the other hand, in practice, modern day neural networks are trained efficiently using SGD and a variety of tricks that include different activation functions (e.g. ReLU), over-specification (i.e., train networks which are larger than needed), a…
Unified framework for training SNNs using EP, faster convergence.
problem Training spiking neural networks with Expectation-Propagation.
method Message-passing framework for learning marginal distributions of SNN parameters.
result Faster convergence compared to gradient-based methods.
Learning portable neural networks is very essential for computer vision for the purpose that pre-trained heavy deep models can be well applied on edge devices such as mobile phones and micro sensors. Most existing deep neural network compression and speed-up methods are very effective for training compact deep models, …
A new training method improves stability and generalization of DeepONets.
problem Training deep operator networks (DeepONets) is challenging due to nonconvex and nonlinear nature.
method Two-step training method: first train trunk network, then branch network. Introduced Gram-Schmidt orthonormalization.
result Generalization error estimate and numerical examples demonstrating effectiveness.
Neural nets trained with linear discriminant initialization converge faster and more accurately.
problem Training feed-forward neural networks efficiently and accurately.
method Initialize first layer weights with linear discriminants.
result Asymptotic higher accuracy and faster convergence.
Proposes dynamic channel pruning during neural network training.
problem Pruning neural networks during training to reduce computational cost and improve efficiency.
method Dynamic channel propagation to update channel utility values and selectively prune channels.
result Our scheme trains and prunes neural networks simultaneously, achieving superior performance.
Ensembles of deep neural networks significantly improve generalization accuracy. However, training neural network ensembles requires a large amount of computational resources and time. State-of-the-art approaches either train all networks from scratch leading to prohibitive training cost that allows only very small ens…
Improved NiNo networks accelerate Adam training by up to 50%.
problem Accelerating neural network training with stable and efficient methods.
method Proposed NiNo networks that leverage neuron connectivity and graph neural networks to nowcast parameters periodically during Adam training.
result Accelerates Adam training by up to 50% in vision and language tasks.
We propose a max-pooling based loss function for training Long Short-Term Memory (LSTM) networks for small-footprint keyword spotting (KWS), with low CPU, memory, and latency requirements. The max-pooling loss training can be further guided by initializing with a cross-entropy loss trained network. A posterior smoothin…
New insights into training deep networks with low rank layers.
problem Efficiency in training deep neural networks.
method Analysis of techniques for training in low rank space.
result Falsified common beliefs in training deep networks.
Proposes faster neural network learning by using subsets of training data.
problem Manual design and validation of network topologies is time-consuming.
method Exploits subsets of training data at each incremental training step and performs online hyperparameter selection.
result Significantly reduces overall training time while maintaining performance.
SingleProp speeds up robust neural network training with minimal certification.
problem Efficiently defending neural networks against adversarial attacks with certified guarantees.
method SingleProp regularizer that requires only one forward pass per training iteration.
result Comparable certified accuracy to state-of-the-art defenses, but significantly faster training.
Deep networks without non-linearities are equivalent to shallow ones.
problem Training deep orthogonal linear networks with no non-linearity.
method Riemannian gradient descent and gradient descent on factorization.
result Training deep overparametrized networks is equivalent to shallow ones.
Parallel neural network training yields better long-term prediction accuracy.
problem Choosing the right training strategy for neural networks in dynamical systems.
method Comparison of parallel and series-parallel training strategies on five neural network architectures and two examples.
result Parallel training consistently outperforms series-parallel training in long-term prediction accuracy.
The paper introduces a multilevel initialization method for deep neural networks.
problem Training very deep neural networks with layer-parallel methods.
method Continuous interpretation of training as optimal control, using time-dependent ODEs for neural network discretization, and a refinement strategy across the time domain.
result The method creates deep networks with good initializations from coarser networks, reducing training time and providing regularization.
Study shows neural networks trained with GD converge to Gaussian processes with polynomial decay.
problem Understanding convergence of neural networks to Gaussian processes during training.
method Explicit upper bounds on quadratic Wasserstein distance between trained networks and Gaussian approximations.
result Polynomial decay of approximation error with network width and training time.
The study finds a theoretical bound for pre-training iterations needed for pruning to yield good subnetwork performance.
problem Discovering efficient subnetworks within pre-trained dense networks.
method Mathematical analysis of a two-layer, fully-connected network, validating with a multi-layer perceptron trained on MNIST.
result A logarithmically dependent threshold on dataset size for successful pruning.
We present DANTE, a novel method for training neural networks using the alternating minimization principle. DANTE provides an alternate perspective to traditional gradient-based backpropagation techniques commonly used to train deep networks. It utilizes an adaptation of quasi-convexity to cast training a neural networ…
The paper trains neural networks with robustness guarantees using semidefinite constraints.
problem Training neural networks with robustness and stability guarantees.
method Exploiting the banded structure of semidefinite constraints, an efficient and scalable training scheme based on interior point methods is set up.
result The method allows for enforcing Lipschitz constraints in large-scale deep neural networks, as demonstrated in numerical examples.
Adjoined Networks trains both base and compressed networks together for efficient model compression.
problem Efficiently compressing deep neural networks while maintaining accuracy.
method Adjoined Networks (AN) trains both a base network and a smaller compressed network simultaneously, sharing parameters.
result AN achieves 71.8% top-1 accuracy with 1.8M parameters and 1.6 GFLOPs on ImageNet.
Deep neural networks have shown remarkable performance across a wide range of vision-based tasks, particularly due to the availability of large-scale datasets for training and better architectures. However, data seen in the real world are often affected by distortions that not accounted for by the training datasets. In…
This paper proposes a new algorithmic framework, predictor-verifier training, to train neural networks that are verifiable, i.e., networks that provably satisfy some desired input-output properties. The key idea is to simultaneously train two networks: a predictor network that performs the task at hand,e.g., predicting…
This work proves composite neural networks outperform their components under certain conditions.
problem Understanding the performance of composite neural networks.
method Theoretical investigation of a composite neural network combining pre-trained models.
result A composite neural network, with high probability, performs better than any of its pre-trained components.
BatchNorm helps train quantized networks by avoiding gradient explosion.
problem Training quantized neural networks is difficult due to gradient issues.
method Investigated the impact of BatchNorm on both full-precision and quantized networks.
result BatchNorm avoids gradient explosion in quantized networks, contrary to expectations.
Study the effects of data parallelism and sparsity on neural network training.
problem Understanding the effects of data parallelism and sparsity on neural network training.
method Conducted extensive experiments and developed a theoretical analysis.
result Found a general scaling trend between batch size and number of training steps to convergence for the effect of data parallelism, and difficulty of training under sparsity.
Deep neural networks with adversarial training achieve sup-norm convergence for nonparametric regression.
problem Achieving sup-norm convergence for deep neural network estimators in nonparametric regression.
method Developed an adversarial training scheme to address the sup-norm convergence issue.
result Deep neural network estimators achieve optimal sup-norm convergence with the proposed adversarial training.
We propose an incremental training method that partitions the original network into sub-networks, which are then gradually incorporated in the running network during the training process. To allow for a smooth dynamic growth of the network, we introduce a look-ahead initialization that outperforms the random initializa…
SMGD trains low-bit neural networks with memory constraints.
problem Training large neural networks with limited memory.
method Stochastic Markov Gradient Descent (SMGD).
result Encouraging numerical results and theoretical guarantees.
Study on neural networks with regularisation and its impact on training dynamics.
problem Understanding the dynamics of neural networks with regularization.
method Established explicit dynamics for neural networks with a regularizing term, linearizing around initialisation.
result The regularisation term modifies the standard NTK dynamics, leading to new insights into network training.
Trained neural networks perform Bayesian reasoning for tasks beyond their initial scope.
problem Performing Bayesian reasoning for tasks outside the trained neural networks' initial scope.
method Used deep generative models as priors and classification/regression networks as constraints. Approximated Bayesian inference through variational or sampling techniques.
result The approach built on top of already trained networks, expanding the addressable questions.
Optical co-processor speeds up neural network training.
problem Expensive training costs for large neural networks.
method Direct feedback alignment, optical error projection.
result Optical co-processor trains neural networks for handwritten digit recognition.
learn2mix trains neural nets faster by adjusting class proportions dynamically.
problem Training neural nets efficiently with limited resources and imbalanced classes.
method Adaptive class proportion adjustment during training.
result Neural nets trained with learn2mix converge faster than static methods.
In this effort we propose a novel approach for reconstructing multivariate functions from training data, by identifying both a suitable network architecture and an initialization using polynomial-based approximations. Training deep neural networks using gradient descent can be interpreted as moving the set of network p…
Sparse routing networks with co-training prevent catastrophic forgetting in continual learning.
problem Catastrophic forgetting in neural networks trained on a sequence of tasks.
method Sparse routing networks with co-training to minimize interference between dissimilar tasks.
result Sparse routing networks with co-training outperform densely connected networks on benchmarks.
Survey on reducing deep neural network training complexity.
problem Efficient training of large deep neural networks.
method Pruning and freezing parts of the network during training.
result Dimensionality reduction improves training efficiency.
Wide stochastic networks show Gaussian behavior and improve training with PAC-Bayesian methods.
problem Analyzing and training over-parameterised neural networks with large width.
method Establishing Gaussian behavior for a stochastic architecture, applying PAC-Bayesian training.
result PAC-Bayesian training on large but finite-width networks outperforms standard methods.
Early neural network training reveals important sub-networks and weight distributions.
problem Understanding the early phases of neural network training.
method Extensive measurements and quantitative probing of weight distribution and dataset reliance.
result Deep networks are not robust to reinitializing with random weights while maintaining signs, and weight distributions are highly non-independent.
High-performance neural network training on a cluster of Pentium III processors.
problem Extremely computationally intensive neural network training.
method Distributed training on a Linux-based cluster of Pentium III processors.
result 1.73 million parameter neural network trained at 163.3 GFLOPS/s.
This paper develops a method to train compact neural networks with reduced memory and computational costs.
problem Training large neural networks consumes excessive resources and energy.
method End-to-end training framework using Bayesian tensor decomposition with automatic rank determination.
result The method achieves significant parameter reduction and maintains or improves accuracy.
The paper analyzes a neural network two-sample test using kernel analysis.
problem Determining if two datasets come from the same distribution.
method Time-analysis on a neural tangent kernel (NTK) two-sample test, extending to realistic neural network dynamics.
result Training times needed to detect deviations are well-separated in null and alternative hypothesis scenarios.