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.

168,742 papers · 148 categories

Trend · papers per month

225449674898 · Jun 202019922001200920172026
48 results for deep diffusion processes

New method DDVI improves posterior inference for deep Gaussian processes.

problem Inference of inducing points in DGPs is challenging and biased.
method DDVI uses denoising diffusion SDE and score matching for posterior approximation.
result Empirically shows DDVI outperforms baseline methods in inducing point inference.

When the parameters are independently and identically distributed (initialized) neural networks exhibit undesirable properties that emerge as the number of layers increases, e.g. a vanishing dependency on the input and a concentration on restrictive families of functions including constant functions. We consider parame…

2019-05-27abs ↗pdf ↗

A new criterion for deep active learning selects minimal labeled data points.

problem Efficiently select minimal labeled data points for deep neural networks.
method Diffuses label information over a graph of data representations to switch between exploration and refinement.
result The diffusion-based criterion outperforms existing methods in deep active learning.

DDP models dynamic comorbidity networks from event data.

problem Understanding complex temporal patterns of co-occurring diseases.
method Developed deep diffusion processes (DDP) to model dynamic comorbidity networks.
result DDP enables accurate risk prediction and interpretable disease trajectories.

Extends RDS filtering to position-orientation space for better image processing.

problem Enhancing and inpainting images with crossing structures.
method Created a version of RDS filtering using gauge frames, studying generalised diffusion.
result RDS filtering on position-orientation space improves denoising and inpainting of crossing structures.

Backpropagation is explained as a diffusion process in neural networks.

problem The biological plausibility of Backpropagation is questioned.
method Demonstrated that time-delayed neurons and forward-backward waves approximate the gradient in deep networks.
result Backpropagation can be interpreted as a diffusion process, approximating the gradient for non-fast inputs.

Diffusion models reveal a phase transition in reconstructing high-level features.

problem Understanding the hierarchical structure of natural data.
method Study of hierarchical generative models of data using diffusion models.
result The backward diffusion process shows a phase transition at a threshold time, where high-level features suddenly drop in reconstructibility.

This paper extends ResNet theory to infinitely deep networks, linking them to diffusion processes.

problem Training infinitely deep ResNets with i.i.d. initializations leads to undesirable properties.
method Introduced doubly infinite ResNets with i.i.d. initializations, linking to diffusion processes.
result The dynamics of quantities of interest converge to deterministic limits in the limit of infinite depth.

Deep learning dynamics exhibit anomalous superdiffusion initially, aiding escape from local minima.

problem Understanding the dynamics of learning in deep neural networks.
method Novel analysis of SGD dynamics and loss landscape structure.
result SGD exhibits anomalous superdiffusion initially, transitioning to subdiffusion as learning progresses.

We study the Stochastic Gradient Descent (SGD) method in nonconvex optimization problems from the point of view of approximating diffusion processes. We prove rigorously that the diffusion process can approximate the SGD algorithm weakly using the weak form of master equation for probability evolution. In the small ste…

2017-05-22abs ↗pdf ↗

A central problem in machine learning involves modeling complex data-sets using highly flexible families of probability distributions in which learning, sampling, inference, and evaluation are still analytically or computationally tractable. Here, we develop an approach that simultaneously achieves both flexibility and…

2015-03-12abs ↗pdf ↗

Automates learning of multivariate diffusions for generative models.

problem Lack of automated methods for choosing and optimizing diffusion processes in generative models.
method Develops a recipe to maximize likelihood without model-specific analysis, parameterizes diffusion for target noise, and optimizes the inference diffusion process.
result Automatic search over all linear diffusions for generative models.

Generative diffusion models mimic biological memory networks, encoding associative dynamics in deep neural weights.

problem Understanding long-term memory mechanisms in neuroscience and AI.
method Interpreting generative diffusion models as energy-based models and comparing them to Hopfield networks.
result Generative diffusion models can encode associative dynamics of Hopfield networks in deep neural weights.

Special issue on understanding physical processes from unusual diffusion patterns.

problem Understanding physical processes from anomalous diffusion data.
method Not explicitly described in the abstract, but likely involves analysis of data from the Anomalous Diffusion Challenge.
result Not explicitly stated, but likely includes analysis of physical processes from anomalous diffusion data.

Efficiently reconstructs jump-diffusion processes from data using neural networks.

problem Reconstructing jump-diffusion processes from data.
method Temporally decoupled squared Wasserstein distance method using parameterized neural networks.
result Enhanced reconstruction of jump-diffusion processes from data.

Masking diffusion outperforms other discrete diffusion models by incorporating jump times into the model.

problem Improving the performance of discrete diffusion models.
method Conditioning on the jump schedule of discrete Markov processes.
result Schedule-conditioned discrete diffusion (SCUD) models outperform classical and masking diffusion models.

Neural Flow Diffusion Models improve diffusion models by learning flexible forward processes.

problem Fixed forward processes in diffusion models complicate reverse processes and increase inference costs.
method Introduces NFDM, a framework supporting flexible forward processes and a novel parameterization technique.
result Demonstrates strong performance in likelihood estimation and learning generative dynamics.

New method improves OOD detection by integrating diffusion models into discriminator models.

problem Overconfidence in discriminator models leads to poor OOD detection.
method Integrates diffusion models into discriminator and generation models to mitigate overconfidence.
result Demonstrates significant improvement in AUROC scores for challenging datasets.

This work connects diffusion models to power iteration, revealing how low frequencies emerge earlier.

problem Understanding the generation process of diffusion models and their relation to power iteration.
method Examined the linear case of diffusion models, connecting them to the spiked covariance model and power iteration.
result Linear diffusion models converge to the leading eigenvector, similar to power iteration.

In this paper, we study the problem of using representation learning to assist information diffusion prediction on graphs. In particular, we aim at estimating the probability of an inactive node to be activated next in a cascade. Despite the success of recent deep learning methods for diffusion, we find that they often…

2017-11-28abs ↗pdf ↗

A new method uses Gaussian processes and deep kernel learning to price high-dimensional American options efficiently.

problem Challenges in pricing high-dimensional American options, especially with excessive computational costs.
method Modified Gaussian process regression with deep kernel learning and sparse variational Gaussian processes.
result The method outperforms least squares Monte Carlo in high-dimensional scenarios, especially with Merton's jump diffusion model.

Rolling Diffusion improves video prediction by progressively corrupting frames based on their temporal position.

problem Improving video prediction accuracy by accounting for temporal dynamics.
method A sliding window denoising process that assigns more noise to frames that appear later in a sequence.
result Rolling Diffusion outperforms standard diffusion models in tasks with complex temporal dynamics.

We study the small-time fluctuations for diffusion processes which are conditioned by their initial and final positions, under the assumptions that the diffusivity has a sub-Riemannian structure and that the drift vector field lies in the span of the sub-Riemannian structure. In the case where the endpoints agree and t…

2016-06-01abs ↗pdf ↗

Extends neural diffusion processes for multi-task regression.

problem Limited to single-task inference, existing formulations cannot capture dependencies across related tasks.
method Introduces a task encoder to condition diffusion model on low-dimensional representations of context observations.
result Improves predictive performance and uncertainty calibration across related functions.

Anomalous diffusion in SGD reveals interactions between hyperparameters and Hessian.

problem Understanding the limiting dynamics of SGD in deep neural networks.
method Continuous-time model of SGD as an underdamped Langevin equation, derived for linear regression.
result Anomalous diffusion is explained by modified loss and probability currents in phase space.

FLDD improves discrete diffusion models by learning a non-Markovian noising process.

problem Efficiency and quality of discrete diffusion models in few-step generation.
method Introduces a learnable non-Markovian forward (noising) process to match the target distribution.
result FLDD produces higher quality samples in fewer steps compared to conventional discrete diffusion models.

A new diffusion model improves time-series forecasting by preserving seasonal patterns.

problem Improving time-series forecasting accuracy, especially for seasonal data.
method A forward diffusion process that decomposes signals into spectral components, altering only the diffusion process.
result The method maintains high signal-to-noise ratios for dominant frequencies, improving long-term pattern recovery.

Study shows how heat leaks from material sets in low diffusivity scenarios.

problem Understanding heat leakage from material sets in low diffusivity limits.
method Generalized leading-order asymptotics for time-dependent diffusion processes.
result Diffusive transport out of a material set is proportional to the surface area of the set boundary.

This work extends Tweedie's formulae to non-Gaussian processes for better diffusion model generation.

problem Limited exploration of non-Gaussian diffusion models and corresponding Tweedie's formulae.
method Extended Tweedie's formulae to geometric Brownian motion, squared Bessel, and Cox-Ingersoll-Ross processes.
result Demonstrated potential of non-Gaussian models in image and financial time series generation.

Deep neural networks can solve optimal stopping problems without dimensionality issues.

problem Optimal stopping problems in high-dimensional state spaces.
method Established a general framework for deep ReLU neural networks to approximate value functions and continuation values.
result Deep neural networks can approximate value functions and continuation values with error at most ε of size κd^q ε^(-r).

Improves generative models by adding jump-diffusion noise.

problem Limited performance of diffusion models in generating samples from unknown distributions.
method Generalizes diffusion processes to include jump-diffusion noise, deriving closed-form generalized score functions.
result Jump-diffusion models outperform Gaussian models in specific parameter regimes.