1-hidden layer networks lead to deep networks that outperform AlexNet on ImageNet.
problem Improving deep neural networks for image classification tasks.
method Sequentially building deep networks layer by layer using shallow networks as auxiliary problems.
result 11-layer network trained by this method exceeds several VGG models on ImageNet.
Analyzes minima of deep linear networks with weight decay.
problem Understanding the loss landscape of deep neural networks.
method Analytical solutions for global minima with weight decay and stochastic neurons.
result The origin is a special point with qualitatively different minima in networks with more than 1 hidden layer.
New proof shows deep neural nets can have sub-optimal local minima.
problem Can over-parameterization eliminate sub-optimal local minima in deep neural networks?
method Counter-example with generic input data and non-linear activation functions.
result Sub-optimal local minima exist in deep neural networks regardless of width.
We propose and analyze a new family of algorithms for training neural networks with ReLU activations. Our algorithms are based on the technique of alternating minimization: estimating the activation patterns of each ReLU for all given samples, interleaved with weight updates via a least-squares step. The main focus of …
We establish upper bounds for the minimal number of hidden units for which a binary stochastic feedforward network with sigmoid activation probabilities and a single hidden layer is a universal approximator of Markov kernels. We show that each possible probabilistic assignment of the states of n output units, given t…
Study Gaussian approximation for deep neural networks with random weights.
problem Understanding the distribution of deep neural networks with random weights.
method Established Gaussian approximation bounds in Wasserstein-1 norm.
result Convergence rates of order n−(1/6)L−1+ε for deep networks with proportional layer widths. New bounds on neural network capacity for treelike sign perceptrons using RDT.
problem Determining the capacity of treelike sign perceptrons neural networks.
method Random Duality Theory (RDT) to establish upper bounds.
result Mathematically rigorous bounds on network capacity for any number of neurons.
We improve recently published results about resources of Restricted Boltzmann Machines (RBM) and Deep Belief Networks (DBN) required to make them Universal Approximators. We show that any distribution p on the set of binary vectors of length n can be arbitrarily well approximated by an RBM with k-1 hidden units, where …
Neural networks favor simple features over complex ones, even when complex features are available.
problem Neural networks exhibit a bias towards simple features over complex ones, even when complex features are present.
method Rigorously defined simplicity bias, theoretical and empirical demonstrations, ensemble approach to improve robustness.
result One hidden layer neural networks favor simple features over complex ones, even in the presence of more robust features.
New insights into Hessian structure of neural networks reveal two forces.
problem Understanding the Hessian structure of neural networks.
method Analyzing the static and dynamic forces, comparing limit distributions using random matrix theory.
result The Hessian structure arises from a combination of static and dynamic forces, with C being a primary driver. Deep networks exhibit permutation saddles and valleys between equivalent minima.
problem Understanding the structure of loss landscapes in deep neural networks.
method Geometric approach to constructing paths between equivalent minima and saddle points.
result Existence of permutation saddles and valleys in deep neural networks.
Almost all local minima in neural networks are strongly convex.
problem The prevalence of strongly convex neighborhoods around local minima in neural network optimization landscapes.
method Rigorous analysis of shallow neural networks with analytic activation functions, dividing parameter space into efficient and redundant domains.
result For shallow neural networks on the efficient domain, almost all local minima are strongly convex.
Improved neural network capacity analysis using simplified RDT.
problem Analyzing the memorization capabilities of sign perceptron neural networks.
method Developed a simplified, partially lifted Random Duality Theory (fl RDT) approach.
result Concrete capacity bounds universally improve over previous best known ones.
Training an artificial neural network involves an optimization process over the landscape defined by the cost (loss) as a function of the network parameters. We explore these landscapes using optimisation tools developed for potential energy landscapes in molecular science. The number of local minima and transition sta…
Paper proves hardness of learning various complex models under local pseudorandom generators.
problem Hardness of learning various complex models.
method Existence of local pseudorandom generators.
result Proves hardness of learning shallow ReLU neural networks and other models.
3-layer NTK models generalize better than 2-layer models, especially with large input dimensions.
problem Understanding the generalization of overparameterized neural networks.
method Analyzing the 3-layer NTK model's test error and comparing it to 2-layer NTK models.
result 3-layer NTK models have a faster descent in test error with respect to the number of neurons in the second hidden layer.
This paper presents two unsupervised learning layers (UL layers) for label-free video analysis: one for fully connected layers, and the other for convolutional ones. The proposed UL layers can play two roles: they can be the cost function layer for providing global training signal; meanwhile they can be added to any re…
Two-layer CNNs can overfit well if initialized correctly.
problem Understanding the conditions for benign overfitting in over-parameterized CNNs.
method Extending analysis to fully trainable two-layer CNNs, examining initialization scaling effects.
result Initialization scaling of the output layer is crucial; large scales lead to fixed output behavior, small scales to complex interactions.
To infer a multilayer representation of high-dimensional count vectors, we propose the Poisson gamma belief network (PGBN) that factorizes each of its layers into the product of a connection weight matrix and the nonnegative real hidden units of the next layer. The PGBN's hidden layers are jointly trained with an upwar…
Network Implosion reduces ResNet layers without accuracy loss.
problem High computation costs in Residual Networks.
method Static layer pruning and retraining to erase unimportant layers.
result Reduces ResNet layers by 24.00-42.86% without accuracy drop.
Proposes a fixed smooth convolutional layer to reduce checkerboard artifacts in CNNs.
problem Checkerboard artifacts in CNNs during upsampling and strided convolution.
method Fixed convolutional layer with adjustable smoothness, applied to four CNNs and GANs.
result Significantly improves classification performance and image generation quality.
Self-attention can replace convolutional layers in vision tasks.
problem The supremacy of convolutional layers in vision tasks.
method Analysis and experiments with self-attention layers compared to convolutional layers.
result Self-attention layers can perform as well as convolutional layers and learn to do so.
Three-layer networks learn more complex features than two-layer networks.
problem Understanding feature learning in deep neural networks.
method Analysis of three-layer neural networks trained with gradient descent.
result Three-layer networks can learn functions that two-layer networks cannot.
To infer multilayer deep representations of high-dimensional discrete and nonnegative real vectors, we propose an augmentable gamma belief network (GBN) that factorizes each of its hidden layers into the product of a sparse connection weight matrix and the nonnegative real hidden units of the next layer. The GBN's hidd…
Study shows LLMs can remove half of layers without significant performance drop.
problem Understanding knowledge storage in LLMs' weights.
method Layer pruning and finetuning to identify and remove unnecessary parameters.
result Minimal degradation of performance after removing up to half of layers.
We improve neural network explainability by bypassing batch normalization.
problem Lack of transparency in neural networks.
method Layer-wise Relevance Propagation with a method to include normalization layers.
result Heatmaps are more accurate for convolutional layers with our method.
MGCN improves multi-layer graph classification using node attributes and relations.
problem Lack of comprehensive multi-layer graph embedding methods considering node attributes and different types of edges.
method Proposes MGCN, a method that combines GCN for multi-layer graphs, incorporating node attributes and both within and between layer relations.
result MGCN outperforms other multi-layer and single-layer methods in semi-supervised node classification tasks.
We address the problem of learning hierarchical deep neural network policies for reinforcement learning. In contrast to methods that explicitly restrict or cripple lower layers of a hierarchy to force them to use higher-level modulating signals, each layer in our framework is trained to directly solve the task, but acq…
A robot learns to classify images with limited perception using a layered reinforcement learning approach.
problem Image classification for robots with partial perception.
method Three-layer architecture using deep reinforcement learning, including meta-layer, action-layer, and classification-layer.
result The method achieves high accuracy on the MNIST dataset and provides explainability of the agent's decision-making process.
Proposes continuous convolution layers for flexible feature map resizing.
problem Fixed stride limitations in discrete convolution layers.
method Introduces Continuous Convolution (CC) layers that use learned continuous functions.
result Dynamic and consistent resizing of feature maps at any scale, non-integer and axis-dependent.
One-layer transformers can't solve induction heads task efficiently.
problem Solving the induction heads task efficiently with one-layer transformers.
method Communication complexity argument showing exponential size requirement.
result No one-layer transformer can solve the induction heads task efficiently.
Unified framework LPCD optimizes quantization of complex submodules.
problem Quantization of complex submodules in neural networks.
method Layer-Projected Coordinate Descent (LPCD) for quantizing arbitrary submodules.
result LPCD enhances both layer-wise PTQ methods and existing submodule approaches.
AIBC uses AI and blockchain for efficient resource sharing.
problem Efficient resource sharing in large-scale decentralized systems.
method Two-consensus scheme with DABFT and DPoEV.
result Predictive deep learning for optimal BFT algorithm selection.
A new neural network model MDRBM improves noise-robustness in classification.
problem Improving noise-robustness in classification problems.
method Stacking a probabilistic-ELM layer on a discriminative restricted Boltzmann machine (DRBM).
result MDRBM outperforms other models, especially in noise-robustness.
Many real-world complex systems across natural, social, and economical domains consist of manifold layers to form multiplex networks. The multiple network layers give rise to nonlinear effect for the emergent dynamics of systems. Especially, weak layers that can potentially play significant role in amplifying the vulne…
New insights suggest larger early layers improve CNN performance.
problem Common assumption of monotonously increasing feature counts in CNNs is challenged.
method Used a skew normal distribution to investigate feature amounts in CNN layers.
result Architectures favoring larger early layers yield better accuracy.
The study shows removing fully connected output layers improves efficiency without sacrificing performance.
problem Large number of parameters in fully connected layers for high-category datasets.
method Examined architectures replacing fully connected output layers with fixed layers and compared performance.
result Fixed classifiers offer no additional benefit over removing the output layer and its parameters.
Our work presents extensive empirical evidence that layer rotation, i.e. the evolution across training of the cosine distance between each layer's weight vector and its initialization, constitutes an impressively consistent indicator of generalization performance. In particular, larger cosine distances between final an…
Unified SVD compression fails in practical tasks, highlighting the importance of per layer activation reconstruction.
problem The failure of a unified SVD compression method in practical tasks like perplexity and accuracy.
method Unified optimization problem for SVD based compression methods, focusing on cross-layer coupling.
result Downstream metrics like perplexity and accuracy degrade severely compared to standard per layer SVD LLM.
Introduces Causal Energy Minimization to understand Transformer layers.
problem Empirical parameterization of Transformer blocks remains largely unexplored.
method Causal Energy Minimization framework that recasts Transformer layers as optimization steps on conditional energy functions.
result Identifies design space for Transformer layers including weight sharing and energy-based interpretations.
The rapid development of high-throughput technologies has enabled the generation of data from biological or disease processes that span multiple layers, like genomic, proteomic or metabolomic data, and further pertain to multiple sources, like disease subtypes or experimental conditions. In this work, we propose a gene…
PLN-Nets with two linear layers and parallel LN achieve universal approximation.
problem Limitations of standard neural network architectures in universal approximation.
method Introduced PLN-Nets combining two linear layers with parallel LN.
result PLN-Nets achieve universal approximation, while standard LN has limited power.
Smectic liquid crystals are materials formed by stacking deformable, fluid layers. Though smectics prefer to have flat, uniformly-spaced layers, boundary conditions can impose curvature on the layers. Since the layer spacing and curvature are intertwined, the problem of finding minimal configurations for the layers bec…
Improved recommendation systems using multi-layer embeddings reduce model size while maintaining accuracy.
problem Improving model accuracy in recommendation systems while minimizing model size.
method Introducing a multi-layer embedding training (MLET) architecture that trains embeddings via a sequence of linear layers.
result Substantial advantages in model accuracy and memory footprint are achieved with reduced embedding dimensions.
Inferencing with network data necessitates the mapping of its nodes into a vector space, where the relationships are preserved. However, with multi-layered networks, where multiple types of relationships exist for the same set of nodes, it is crucial to exploit the information shared between layers, in addition to the …
New approach replaces BN layers with shifted-ReLU for embedded systems.
problem Complexity and overheads of BN layers in low-power embedded systems.
method Used shifted-ReLU layers instead of BN layers in wide residual networks.
result Shifted-ReLU layers offer advantages in speed, memory, and complexity without significant accuracy loss.
A method to analyze neural network performance by measuring layer saturation.
problem Understanding which layers contribute to network performance.
method Layer saturation method: restricts layer output to eigenspace of variance matrix.
result Layer saturation indicates which layers contribute to network performance.
A new tensor-based layer reduces neural network dimensions without losing important features.
problem Reducing dimensionality in tensor-structured feature data for deep neural networks.
method TensorProjection layer that projects input tensors into output tensors with reduced dimensions through mode-wise projections.
result The TensorProjection layer outperforms traditional downsampling methods in tasks like medical image classification and segmentation.