Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

169,341 papers · 148 categories

Trend · papers per month

2.4%4.8%7.1%9.5% · Sep 199619922001200920182026
48 results for alternating back-propagation

Paper proposes an alternating back-propagation algorithm for generator networks.

problem Learning realistic generator models of natural images, video sequences, and sounds.
method Alternating back-propagation algorithm that iterates inferential and learning steps.
result The alternating back-propagation algorithm can learn realistic generator models of natural images, video sequences, and sounds.

Paper proposes new neural network learning algorithms inspired by predictive coding.

problem Finding biologically plausible alternatives to back-propagation of errors.
method Error-driven Local Representation Alignment (LRA-E) and Difference Target Propagation.
result Both proposed algorithms yield stable performance and strong generalization in training deeper, highly nonlinear networks.

Paper learns dynamic generator models for video sequences.

problem Modeling spatial-temporal processes like dynamic textures and actions.
method Alternating back-propagation through time algorithm to learn latent state vectors and generator model.
result Trains realistic models for dynamic textures and actions.

Functional transfer matrices replace weights in neural networks, achieving high accuracy.

problem Representing connections in neural networks with functions instead of weights.
method Developed functional transfer matrices, stacked them with bias vectors and activations, and trained them using back-propagation.
result Deep functional transfer neural networks can be trained to achieve high test accuracies on the MNIST database.

ProxQuant improves quantized neural networks using proximal operators.

problem Making neural networks work on devices with limited resources.
method Formulates quantized network training as a regularized learning problem and optimizes it via the prox-gradient method.
result ProxQuant outperforms state-of-the-art results on binary quantization and is on par with state-of-the-art on multi-bit quantization.

Truncated back-propagation improves hyperparameter tuning and meta learning efficiency.

problem Computational challenges in evaluating exact gradients for high-dimensional bilevel optimization problems.
method Use truncated back-propagation to approximate gradients for the lower-level problem.
result Optimization with few-step back-propagation approximations often performs comparably to exact gradients, but with less memory and computation.

LRA trains deep networks robustly with less sensitivity to initial weights.

problem Training deep networks is challenging due to issues like exploding and vanishing gradients.
method Local Representation Alignment (LRA) is a training procedure less sensitive to initial weights.
result LRA can train networks robustly, even with null initial weights, and outperforms other methods.

The back-propagation algorithm is widely used for learning in artificial neural networks. A challenge in machine learning is to create models that generalize to new data samples not seen in the training data. Recently, a common flaw in several machine learning algorithms was discovered: small perturbations added to the…

2015-10-14abs ↗pdf ↗

A new method reduces backpropagation in CNNs by passing only a small subset of gradients.

problem High computational resources in traditional CNN backpropagation.
method Minimal Effort Back Propagation (MEBP) for CNNs, selecting only a small subset of gradients.
result Only 5% of gradients are passed back, achieving the same or better results as traditional CNNs.

Proposes a semi-implicit back propagation method for neural networks.

problem Challenges in training neural networks, especially gradient vanishing and small step sizes.
method Proposes a semi-implicit back propagation method using error back propagation and proximal methods.
result The proposed method leads to better performance in terms of loss decreasing and training/validation accuracy compared to SGD and ProxBP.

Deep networks trained with Hebbian updates perform similarly to back-propagation on image datasets.

problem Training deep networks with realistic asymmetric connections and updates.
method Use Hebbian updates with separate feedforward and feedback weights, and local rule for updates.
result Similar performance to back-propagation achieved with Hebbian updates on challenging image datasets.

The paper introduces various gradient descent algorithms for training deep learning models.

problem Training deep neural networks is challenging due to their complexity.
method Gradient descent and its variants are discussed for optimizing deep learning models.
result Gradient descent and its variants improve the training performance of deep learning models.

The study learns neural update rules by remembering past experiences.

problem Developing efficient online learning rules for neural networks.
method Representing neurons with vectors, using meta-neural networks for updates, and training for remembering past experiences.
result The approach reveals insights into learning rules and could be used for complex tasks like episodic memory.

NoProp learns neural networks without full back-propagation or forward-propagation.

problem Learning hierarchical representations in neural networks.
method NoProp independently learns each block to denoise a noisy target using local targets and back-propagation within the block.
result NoProp is a viable learning algorithm that is easy to use and computationally efficient.

A simplified backpropagation method reduces model complexity and computational cost.

problem Complexity and computational inefficiency in deep learning models.
method Sparsification of gradient vectors and adaptive model simplification.
result Models can be simplified by updating only a small fraction of weights, improving accuracy.

Paper uses AI to predict stock market volatility with neural networks and genetic algorithms.

problem Traditional methods for predicting stock market volatility have high errors.
method Back-propagation neural network and genetic algorithm integrated model.
result The model predicts future volatility with low errors and high accuracy.

Defines a calculus for integrating Moreau envelopes in differentiable programming.

problem Lack of a mathematical framework for applying Moreau envelopes to deep networks and machine learning systems.
method Develops a compositional calculus adapted to Moreau envelopes and integrates it into differentiable programming.
result Integrates Moreau envelopes into differentiable programming, enabling new gradient back-propagation methods.

A new method improves fitting neural data with spiking network models.

problem Fitting spiking network models to neural activity does not produce realistic data.
method Augment log-likelihood with dissimilarity terms measured by summary statistics and optimized via back-propagation.
result The new method generates more realistic neural activity statistics and improves network connectivity inference.

Optimization-based pruning eliminates backpropagation for large language models.

problem Suboptimal pruning performance due to heuristic metrics.
method Optimization of Bernoulli distribution to learn pruning masks without backpropagation.
result Efficient pruning of large language models with improved performance.

EXODUS improves training of SNNs by stabilizing gradients and reducing complexity.

problem Training SNNs using BPTT is time-consuming and numerically unstable.
method EXODUS modifies SLAYER to account for neuron reset and uses IFT for correct gradient calculation, eliminating manual scaling.
result EXODUS achieves comparable or better performance than SLAYER, especially in tasks with temporal features.

Paper justifies ST estimator using pWGF and proposes an improved variant.

problem Theoretical justification for ST estimator for discrete variables.
method Interpreted ST as pWGF simulation and proposed an improved estimator.
result Established theoretical foundation for ST estimator and improved variant.

LDR neural networks reduce space and complexity with high accuracy.

problem Reducing space and computational complexity in large-scale neural networks.
method Formal study of LDR matrices, proving approximation property, error bounds, and proposing training algorithm.
result LDR neural networks achieve high accuracy with significant reduction in space and computational complexity.

FrequentNet uses frequency domain basis vectors for image classification, making models more interpretable and efficient.

problem Image classification models are often complex and hard to interpret.
method FrequentNet selects filter vectors from frequency domain basis vectors instead of training them with back propagation.
result The method improves interpretability and efficiency of image classification models.

PLIs improve classifier performance by fine-tuning latent representations.

problem Difficult interpretation of high-dimensional latent representations in neural networks.
method Back-propagation of manual changes to low-dimensional embeddings using t-distributed stochastic neighbourhood embeddings.
result Manual separation of class clusters in latent space enhances classifier performance.

A new probabilistic framework improves deep learning performance.

problem Improving deep learning models and understanding their limitations.
method Developed a probabilistic framework based on DRMM, a generative model capturing latent variables.
result DRMM outperforms DCNs in classification tasks, achieving state-of-the-art results.

PredRNN++ improves video predictive learning by alleviating gradient issues.

problem Gradient vanishing problem in deep spatiotemporal models.
method Causal LSTM and Gradient Highway architecture to manage gradients.
result PredRNN++ yields state-of-the-art prediction results even in occlusion scenarios.

A new gradient method reduces variance for non-reparameterizable distributions.

problem Efficient calculation of unbiased gradients for expectation-based objectives.
method GO Gradient, which applies to non-reparameterizable distributions and has low variance.
result GO Gradient reduces variance to the same level as reparameterization trick with one sample.

This paper proposes an alternative to E2E training for deep networks, reducing memory footprint.

problem High GPUs memory footprint in end-to-end training of deep networks.
method Locally supervised learning with information propagation loss to avoid information collapse.
result The proposed method achieves competitive performance with less than 40% memory footprint compared to E2E training.

Paper visualizes and analyzes deep neural networks for CTR prediction.

problem Interpreting deep neural networks for CTR prediction is challenging.
method Visualizing and analyzing deep neural networks at neuron level, measuring layer-wise performance, and calculating saliency scores.
result Insights into the model's performance and mechanism.