New layer sparsity concept improves neural networks.
problem Improving neural network efficiency and interpretability.
method Formulated layer sparsity, introduced regularization and refitting schemes.
result Generated more compact and accurate neural networks.
Various forms of representations may arise in the many layers embedded in deep neural networks (DNNs). Of these, where can we find the most compact representation? We propose to use a pruning framework to answer this question: How compact can each layer be compressed, without losing performance? Most of the existing DN…
LEWIS merges LLMs without training, improving performance on specific tasks.
problem Limited performance improvement of merged models on specific benchmarks.
method Guided model merging using layer-wise sparsity and task-vector pruning.
result Improved model performance by up to 11.3% on math-solving tasks.
DSA efficiently allocates sparsity across layers for budgeted pruning.
problem Efficiently distributing resources (sparsity) across layers in pruning under resource constraints.
method DSA uses differentiable pruning to find continuous layer-wise pruning ratios via gradient-based optimization.
result DSA achieves superior performance and significantly reduces the time cost of pruning.
High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) l…
New method targets sparsity to prevent overfitting in deep nets.
problem Overfitting in deep neural networks with small datasets.
method Targeted sparsity regularization to visualize and counteract overfitting.
result Significant increase in image classification performance without overfitting.
Grassmannian packings improve CNN kernels' diversity and reduce sparsity.
problem Kernel sparsity and lack of diversity in CNNs decrease model capacity.
method Initialize CNN kernels with Grassmannian packings to maximize diversity and minimize sparsity.
result Grassmannian packings lead to diverse features and improved classification accuracy.
Sparse activations in neural networks are hard to exploit but lead to advantages in learning.
problem Sparse activations in neural networks are hard to exploit but lead to advantages in learning.
method Formal study of PAC learnability of MLP layers with activation sparsity.
result Classes of functions with activation sparsity lead to provable computational and statistical advantages over their non-sparse counterparts.
2L-FUSE enhances feature sparsity through kernel learning.
problem Sparsity and feature selection in regression tasks.
method 2-Layered kernel machines for learning a shape matrix and feature direction identification.
result Minimal yet informative feature sets are identified without losing predictive performance.
This paper accelerates sparse CNN layers on GPUs by using unstructured sparsity.
problem Efficiency of sparse CNN layers on GPUs.
method Direct sparse operation and reduced precision.
result Achieving up to 90% sparsity in deep CNN models improves efficiency.
MoEs can efficiently model complex tasks with low-dimensionality and sparsity.
problem Understanding the theoretical foundations of MoEs for complex tasks.
method Systematic study of MoEs with two structural priors: low-dimensionality and sparsity.
result MoEs can approximate functions on low-dimensional manifolds and exhibit exponential structured tasks.
Transformers exhibit sparse activation maps, reducing computational load and improving robustness.
problem Sparse activation in Transformer models.
method Extensive experiments on various Transformer architectures and tasks.
result Sparsity in Transformers is a prevalent phenomenon, reducing FLOP count and improving model robustness.
Polynomial-time convex optimization for CNNs with ReLU activations.
problem Training Convolutional Neural Networks (CNNs) with ReLU activations.
method Developed a convex analytic framework using semi-infinite duality to formulate equivalent convex optimization problems for CNN architectures.
result Proved that two-layer CNNs can be globally optimized via an ℓ2 norm regularized convex program. Sparse Transformers can approximate dense Transformers with only O(n) connections.
problem Can sparse Transformers approximate arbitrary sequence-to-sequence functions?
method Proposed sufficient conditions for universal approximation and proved that sparse Transformers with O(n) connections can approximate dense models.
result Sparse Transformers with O(n) connections can approximate the same function class as dense models with n^2 connections.
Attention mechanisms have become ubiquitous in NLP. Recent architectures, notably the Transformer, learn powerful context-aware word representations through layered, multi-headed attention. The multiple heads learn diverse types of word relationships. However, with standard softmax attention, all attention heads are de…
BiSHop tackles tabular data challenges with sparse Hopfield layers.
problem Non-rotationally invariant data structure and feature sparsity in tabular data.
method Sequential column-wise and row-wise processing through interconnected directional learning modules with generalized sparse modern Hopfield layers.
result BiSHop surpasses current SOTA methods with significantly less hyperparameter tuning.
New model for multi-layer categorical data improves latent class analysis.
problem Traditional latent class analysis for single-layer categorical data is insufficient for multi-layer data.
method Developed a multi-layer latent class model (multi-layer LCM) and three spectral methods for estimation.
result The debiased sum of Gram matrices method performs best in estimating latent classes.
The mushroom body is the key network for the representation of learned olfactory stimuli in Drosophila and insects. The sparse activity of Kenyon cells, the principal neurons in the mushroom body, plays a key role in the learned classification of different odours. In the specific case of the fruit fly, the sparseness o…
STR reparameterizes DNN weights with soft thresholds for better sparsity and accuracy.
problem Improving sparsity in DNNs for better accuracy and lower inference cost.
method Soft Threshold Reparameterization (STR) using the soft-threshold operator on DNN weights.
result STR achieves state-of-the-art accuracy and reduces FLOPs by up to 50%.
Learned data models based on sparsity are widely used in signal processing and imaging applications. A variety of methods for learning synthesis dictionaries, sparsifying transforms, etc., have been proposed in recent years, often imposing useful structures or properties on the models. In this work, we focus on sparsif…
We establish L∞ and L2 error bounds for functions of many variables that are approximated by linear combinations of ReLU (rectified linear unit) and squared ReLU ridge functions with ℓ1 and ℓ0 controls on their inner and outer parameters. With the squared ReLU ridge function, we show th…
SparseRT accelerates sparse computations on GPUs for deep learning inference.
problem Efficiently handling unstructured sparsity patterns on GPUs for deep learning.
method SparseRT, a code generator that leverages unstructured sparsity for accelerating sparse linear algebra operations.
result Geometric mean speedups of 3.4x at 90% sparsity and 5.4x at 95% sparsity for 1x1 convolutions and fully connected layers.
NFM improves deep learning by selectively processing hidden states.
problem Processing entire hidden states in each layer limits modularity and reusability.
method Introduces Neural Function Modules (NFM) with attention, sparsity, and feedback.
result Improves results in classification, generalization, generative modeling, and reinforcement learning.
New sparse penalty improves biclustering for gene expression data.
problem Identifying significant clusters in gene expression data.
method Prenet penalty applied to SSVD for biclustering.
result Mixed Prenet penalty effectively clusters non-overlapped data.
Despite their impressive performance, deep convolutional neural networks (CNNs) have been shown to be sensitive to small adversarial perturbations. These nuisances, which one can barely notice, are powerful enough to fool sophisticated and well performing classifiers, leading to ridiculous misclassification results. In…
This paper proposes an automatic neural network compression method.
problem Reducing resource requirements for deep neural networks on resource-constrained devices.
method Jointly prunes and quantizes neural networks without manual hyper-parameter tuning.
result Significant reduction in model size with minimal accuracy loss.
We study the emergence of sparse representations in neural networks. We show that in unsupervised models with regularization, the emergence of sparsity is the result of the input data samples being distributed along highly non-linear or discontinuous manifold. We also derive a similar argument for discriminatively trai…
New algorithm finds important synapses without training data.
problem Finding important synapses in neural networks without data.
method Iterative Synaptic Flow Pruning (SynFlow) based on conservation law.
result Algorithm consistently outperforms existing pruning methods.
This paper investigates the role of sparsity in Reservoir Computing networks.
problem Designing efficient Recurrent Neural Networks (RNNs) with hidden recurrent layers.
method Empirical investigation of sparsity in input-reservoir connections and recurrent connections.
result Sparsity, particularly in input-reservoir connections, enhances the network's temporal memory and dimensionality.
SparseTrain uses dynamic sparsity in training deep neural networks on CPUs.
problem Training deep neural networks efficiently on general-purpose processors.
method Exploits dynamic zeros introduced by ReLU in feature maps and gradients.
result Significantly speeds up training on CPUs, up to 1.51x.
Lookahead pruning extends single-layer optimization to multi-layer, outperforming magnitude-based pruning.
problem Pruning neural networks to reduce computational cost and memory usage.
method Developed a multi-layer optimization approach extending the single-layer optimization of magnitude-based pruning.
result Consistently outperforms magnitude-based pruning on various networks, especially in high sparsity.
A new framework for interpretable models using sparse linear layers.
problem Performance degradation and lower interpretability in concept bottleneck models.
method Contrastive Language Image models and a single sparse linear layer with Bayesian inference.
result Our framework outperforms recent CBM approaches in accuracy and concept sparsity.
DNNs with L2 regularization reveal feature learning dynamics and sparsity.
problem Understanding feature learning in DNNs with L2 regularization. method Reformulating loss in terms of layerwise activations and covariances.
result Proving sparsity of local minima in L2-regularized DNNs. DeepLight accelerates CTR predictions in ad serving by 46X.
problem Significantly increased serving delay and high memory usage for ad serving.
method Explicitly searching feature interactions, pruning layers, promoting sparsity.
result Accelerates model inference by 46X on Criteo dataset.
MLP-Mixer achieves better performance through sparsity and wider architecture.
problem Understanding why MLP-Mixer outperforms conventional MLPs.
method Revealed sparseness as a key mechanism, showed effective expression as wider MLP, demonstrated quantitative similarities, and applied a guiding principle.
result MLP-Mixer's performance improvement through sparseness and wider architecture.
Paper proposes efficient pruning method for neural networks.
problem Compressing deep neural networks for resource-constrained devices.
method Adaptive sparsity loss for budget-aware optimization during training.
result Demonstrated effectiveness on various architectures and datasets.
Transformers can learn optimal variable selection in group-sparse classification.
problem Understanding how transformers leverage attention to select relevant variables in group-sparse classification.
method Training a one-layer transformer using gradient descent to select variables from one group of input variables.
result A one-layer transformer can correctly leverage the attention mechanism to select variables, disregarding irrelevant ones.
AlphaPruning optimizes LLM pruning using HT-SR theory for better performance.
problem Improving pruning of large language models to reduce size without sacrificing performance.
method AlphaPruning uses HT-SR theory to allocate layerwise sparsity ratios more theoretically.
result AlphaPruning prunes LLaMA-7B to 80% sparsity with reasonable perplexity.
Signal models based on sparsity, low-rank and other properties have been exploited for image reconstruction from limited and corrupted data in medical imaging and other computational imaging applications. In particular, sparsifying transform models have shown promise in various applications, and offer numerous advantag…
Autoencoders fail to capture sparse structure in 1-bit data compression.
problem Proving the performance of shallow autoencoders on sparse data compression.
method Gradient descent analysis and approximate message passing.
result Gradient descent minimizer for sparse data is the identity (up to permutation) above critical sparsity.
We investigate the problem of factorizing a matrix into several sparse matrices and propose an algorithm for this under randomness and sparsity assumptions. This problem can be viewed as a simplification of the deep learning problem where finding a factorization corresponds to finding edges in different layers and valu…
MuVI models multi-view data with structured sparsity, integrating domain knowledge.
problem Disentangling variation across multiple data views in complex systems.
method Multi-view latent variable model with structured sparsity using a modified horseshoe prior.
result MuVI outperforms state-of-the-art methods in structured sparsity modeling and integrates noisy domain expertise.
Proposes a co-hub node model for multiview graph learning.
problem Identifying shared graphical structures in heterogeneous datasets.
method Enforces structured sparsity on co-hub nodes across multiple views.
result Demonstrates improved precision and interpretive insight in learning multiview graphs.
This paper explores loss landscapes of sparse neural networks, finding unique characteristics compared to dense networks.
problem Understanding the loss landscape of sparse neural networks, especially one-hidden-layer networks.
method Analyzes sparse networks with dense and sparse final layers, focusing on linear and non-linear models.
result Sparse networks can have no spurious valleys under certain conditions, but spurious valleys and minima can exist for wide sparse networks.
This work extends neural networks to automatically select features by stochastically penalizing feature involvement.
problem Feature selection in machine learning models.
method Stochastic regularization to select features instead of layer weights.
result Superior efficiency compared to classical methods with minimal computational overhead.
Feature selection is a dimensionality reduction technique that selects a subset of representative features from high dimensional data by eliminating irrelevant and redundant features. Recently, feature selection combined with sparse learning has attracted significant attention due to its outstanding performance compare…
This paper proposes a novel Stochastic Split Linearized Bregman Iteration (S2-LBI) algorithm to efficiently train the deep network. The S2-LBI introduces an iterative regularization path with structural sparsity. Our S2-LBI combines the computational efficiency of the LBI, and model selection consistency…
ResRep prunes CNNs without losing accuracy by separating remembering and forgetting.
problem Pruning CNNs to reduce FLOPs without sacrificing accuracy.
method Decoupling remembering and forgetting in CNNs, using SGD for remembering and a novel update rule for forgetting.
result Achieved lossless pruning with high compression ratio (76.15% accuracy on ImageNet with 45% FLOPs reduction).