There is a general trend towards solving problems suited to deep learning with more complex deep learning architectures trained on larger training sets. This requires longer compute times and greater data parallelization or model parallelization. Both data and model parallelism have been historically faster in paramete…
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.
168,695 papers · 148 categories
Trend · papers per month
48 results for “dynamic architecture”
Dynamic Sparse Training: Find Efficient Sparse Network From Scratch With Trainable Masked Layerscs.LG
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.
dynoNet learns dynamical systems using linear operators.
problem Learning complex dynamical systems.
method dynoNet uses linear dynamical operators for sequence modeling and system identification.
result dynoNet effectively identifies systems on benchmarks.
Capsule networks improve with dynamic routing using Wasserstein objective.
problem Capsule networks struggle to consistently outperform traditional neural networks.
method Dynamic routing scheme using approximate Wasserstein objective to select capsules.
result Capsule network achieves over 1.2% improvement on CIFAR-10 with fewer parameters.
New method grows deep networks efficiently by dynamically pruning and growing layers.
problem Training deep networks is computationally expensive and inefficient.
method Structured continuous sparsification starting from a small seed architecture.
result 49.7% inference FLOPs and 47.4% training FLOPs savings with 75.2% top-1 accuracy.
This work introduces the concept of tangent space regularization for neural-network models of dynamical systems. The tangent space to the dynamics function of many physical systems of interest in control applications exhibits useful properties, e.g., smoothness, motivating regularization of the model Jacobian along sys…
In a recurrent setting, conventional approaches to neural architecture search find and fix a general model for all data samples and time steps. We propose a novel algorithm that can dynamically search for the structure of cells in a recurrent neural network model. Based on a combination of recurrent and recursive neura…
The paper explores how structured representations influence learning dynamics in neural networks.
problem Understanding the training dynamics of deep neural networks.
method Investigates a family of enriched transformation layers with constrained pathways and adaptive corrections.
result Improved robustness, smoother optimization, and scalable depth behavior are achieved through structured representations.
Learning workable representations of dynamical systems is becoming an increasingly important problem in a number of application areas. By leveraging recent work connecting deep neural networks to systems of differential equations, we propose \emph{variational integrator networks}, a class of neural network architecture…
GD-VAEs learn dynamics from observations using geometric and topological information.
problem Learning parsimonious representations of nonlinear dynamics from observations.
method Develops data-driven methods incorporating geometric and topological information using Variational Autoencoders (VAEs).
result GD-VAEs provide methods for learning reduced dimensional representations of nonlinear dynamics.
Advanced simulation paradigm of human behaviour unveils complex financial systemic projectionq-fin.TR
New simulation model predicts financial market dynamics with high accuracy.
problem Extreme difficulty in financial market projections due to human behavioural complexity.
method Agent-based modeling with a hierarchical knowledge architecture to simulate diverse human groups.
result Simulator achieves 13.29% deviation in crisis scenarios and lower mean square error under normal conditions.
Neural Network is a powerful Machine Learning tool that shows outstanding performance in Computer Vision, Natural Language Processing, and Artificial Intelligence. In particular, recently proposed ResNet architecture and its modifications produce state-of-the-art results in image classification problems. ResNet and mos…
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.
Improved SINDy autoencoder for identifying noisy dynamical systems.
problem Robust identification of noisy dynamical systems from data.
method Incorporates noise-separating neural network structures into SINDy autoencoder architecture.
result Accurately recovers latent dynamics and estimates measurement noise from noisy observations.
New neural network designs learn contact dynamics efficiently.
problem Learning contact dynamics in robotics from noisy data.
method Physically structured neural networks.
result Data-efficient learning of discontinuous contact events.
Improved recurrent neural networks learn long-term dependencies through multi-scale memory.
problem Capturing long-term dependencies in recurrent neural networks.
method Incremental training of a modular RNN architecture with multi-scale hidden states.
result Incremental training and multi-scale memory enhance RNNs' ability to learn long-term dependencies.
DyNODE uses neural ODEs to model system dynamics in continuous control tasks.
problem Modeling the dynamics of systems in continuous control tasks.
method Neural Ordinary Differential Equations (ODEs) combined with actor-critic RL.
result DyNODE outperforms standard neural networks in sample efficiency and predictive performance.
AdaEnsemble learns adaptive feature interactions for CTR prediction.
problem Learning feature interactions for CTR prediction in recommender systems and Ads ranking.
method AdaEnsemble is a Sparsely-Gated Mixture-of-Experts (SparseMoE) architecture that dynamically selects feature interaction depth.
result AdaEnsemble achieves better prediction accuracy and inference efficiency compared to state-of-the-art models.
We develop a deep autoencoder architecture that can be used to find a coordinate transformation which turns a nonlinear PDE into a linear PDE. Our architecture is motivated by the linearizing transformations provided by the Cole-Hopf transform for Burgers equation and the inverse scattering transform for completely int…
Lyapunov analysis improves RNN performance prediction.
problem Uncertainty in RNN performance prediction due to hyperparameters and architecture.
method Lyapunov spectral analysis of RNNs and Autoencoder-Lyapunov Embedding Learning (AeLLE).
result AeLLE successfully correlates RNN Lyapunov spectrum with accuracy and predicts performance.
Object Files and Schemata: Factorizing Declarative and Procedural Knowledge in Dynamical Systemscs.LG
New architecture separates object state and behavior for better game dynamics.
problem Lack of systematicity in black-box models for dynamic environments.
method Factorizes declarative and procedural knowledge into object files and schemata.
result Improves generalization on environments with multiple object tokens.
Proposes a Koopman operator method for time-dependent reliability analysis of nonlinear systems.
problem Challenges in time-dependent reliability analysis of nonlinear dynamical systems.
method Koopman operator approach for transforming nonlinear systems into linear ones, combined with deep learning for intrinsic coordinates.
result Robust and generalizable approach for time-dependent reliability analysis, superior to purely data-driven methods.
EvoNet predicts the evolution of dynamic graphs using a graph neural network and recurrent architecture.
problem Predicting the evolution of dynamic graphs is challenging and underexplored.
method EvoNet uses a graph neural network and recurrent architecture to predict the evolution of dynamic graphs.
result EvoNet effectively predicts the evolution of dynamic graphs on both artificial and real-world datasets.
Latent MoS learns multiple symmetries for efficient dynamic learning.
problem Efficiently learning dynamics from limited system measurements.
method Latent Mixture of Symmetries (Latent MoS) with hierarchical architecture.
result Latent MoS outperforms baselines in interpolation and extrapolation tasks.
New neural net learns time-reversible symplectic dynamics.
problem Lack of time-reversibility in neural networks for symplectic systems.
method Proposes a new neural network architecture for time-reversible symplectic systems.
result Demonstrates learning of time-reversible symplectic dynamics from data.
The deployment of deep neural networks in real-world applications is mostly restricted by their high inference costs. Extensive efforts have been made to improve the accuracy with expert-designed or algorithm-searched architectures. However, the incremental improvement is typically achieved with increasingly more expen…
Learning Latent Graph Geometry via Fixed-Point Schrödinger-Type Activation: A Theoretical Studycs.LG
Study neural architectures on learned latent graphs using Schrödinger dynamics.
problem Understanding neural architectures on learned latent graphs.
method Optimizes over stratified moduli space of weighted graphs with Kähler-Hessian metric.
result Multilayer stationary networks are equivalent to global stationary problems on supra-graphs.
Develops theory for data-driven methods in dynamical systems.
problem Lack of analysis for data-driven methods in dynamical systems.
method Establishes existence of mapping and properties of operator learning architecture.
result Novel universal approximation theorems for smoothing and forecasting.
Modeling brain dynamics to better understand and control complex behaviors underlying various cognitive brain functions are of interests to engineers, mathematicians, and physicists from the last several decades. With a motivation of developing computationally efficient models of brain dynamics to use in designing cont…
Proposes a new neural network architecture combining MLP and basis functions.
problem Function approximation and operator learning in scientific machine learning.
method Combines robust MLP inner functions with flexible basis functions outer functions.
result KKAN outperforms MLPs and KANs in function approximation and operator learning tasks.
State-of-the-art named entity recognition (NER) systems have been improving continuously using neural architectures over the past several years. However, many tasks including NER require large sets of annotated data to achieve such performance. In particular, we focus on NER from clinical notes, which is one of the mos…
Recurrent neural networks (RNNs) are powerful dynamical models for data with complex temporal structure. However, training RNNs has traditionally proved challenging due to exploding or vanishing of gradients. RNN models such as LSTMs and GRUs (and their variants) significantly mitigate these issues associated with trai…
Inverse depth scaling found in LLMs due to similar layers averaging error.
problem Understanding how depth affects loss in large language models.
method Analysis of LLMs and toy residual networks.
result Loss scales inversely proportional to depth in LLMs.
New scaling framework for MoE architectures ensures stability and optimal performance at scale.
problem Lack of principled understanding of how hyperparameters should scale in MoE architectures.
method Developed a novel Dynamical Mean Field Theory (DMFT) for three scaling regimes of MoE architectures.
result Derived Maximally Scale-Stable Parameterization (MSSP) for SGD and Adam, providing robust learning rate transfer and monotonic improvement with scale.
Evaluation of the MACE Force Field Architecture: from Medicinal Chemistry to Materials Sciencephysics.chem-ph
MACE architecture outperforms alternatives in various molecular and materials science tasks.
problem Improving machine learning force fields for diverse molecular and materials science applications.
method Evaluation of MACE architecture on various datasets and tasks, demonstrating data efficiency and excellent performance.
result MACE architecture generally outperforms alternatives across a wide range of systems, including amorphous carbon, universal materials modelling, and organic chemistry.
Proposes a new training algorithm for zero-sum games to avoid convergence issues.
problem Gradient-based training leads to weak convergence and cyclic dynamics in zero-sum architectures.
method Follow the perturbed leader algorithm with neural mediating agent.
result Guarantees convergence to mixed Nash equilibrium without cyclic behaviors.
New model predicts neural network performance from early training epochs, incorporating architecture impact.
problem Predicting neural network performance from early training epochs, neglecting architecture impact.
method Architecture-aware graph ordinary differential equation model.
result Model outperforms state-of-the-art methods for MLP and CNN learning curves.
STRIC detects anomalies in time series by analyzing residual signals.
problem Anomaly detection in multivariate time series data.
method End-to-end differentiable neural network architecture with Sequential Probability Ratio Test on residuals.
result STRIC outperforms state-of-the-art methods on multiple benchmarks.
We introduce Dynamic Planning Networks (DPN), a novel architecture for deep reinforcement learning, that combines model-based and model-free aspects for online planning. Our architecture learns to dynamically construct plans using a learned state-transition model by selecting and traversing between simulated states and…
Financial markets modeled like brain networks using dMNC.
problem Understanding latent dynamics in financial markets.
method Biologically inspired framework using dMNC.
result Structural persistence, regime shifts, and early warning signals identified.
Researchers dissect Neural ODEs to understand their dynamics.
problem Understanding the inner workings of Neural ODEs.
method Developing continuous-depth formulation to clarify design choices.
result Clarified the influence of design choices on Neural ODE dynamics.
SympFormer accelerates attention blocks using inertial dynamics on density spaces.
problem Improving the efficiency of self-attention blocks in Transformers.
method Introduced accelerated attention blocks derived from inertial Nesterov dynamics on density spaces.
result Accelerated attention blocks converge faster than classical blocks while preserving oracle calls.
This paper introduces a novel optimization method for differential neural architecture search, based on the theory of prediction with expert advice. Its optimization criterion is well fitted for an architecture-selection, i.e., it minimizes the regret incurred by a sub-optimal selection of operations. Unlike previous s…
Dissipative SymODEN: Encoding Hamiltonian Dynamics with Dissipation and Control into Deep Learningcs.LG
Dissipative SymODEN learns dynamics with dissipation and control from data.
problem Learning dynamics with dissipation and control from observed data.
method Dissipative SymODEN encodes port-Hamiltonian dynamics into a deep learning architecture.
result The learned model reveals key aspects of the system, such as inertia, dissipation, and potential energy.
Collegial Ensemblescs.LG
Ensembles of neural networks improve training dynamics and performance.
problem Improving neural network performance through model size increase.
method Defining collegial ensembles (CE) as multiple independent models trained as a single model, and using theoretical results on NTK to optimize architecture search.
result CE dynamics simplify and scale favorably, resembling wide models, and can be efficiently implemented using group convolutions and block diagonal layers.
Recurrent and convolutional neural networks are the most common architectures used for time series forecasting in deep learning literature. These networks use parameter sharing by repeating a set of fixed architectures with fixed parameters over time or space. The result is that the overall architecture is time-invaria…
A main theoretical interest in biology and physics is to identify the nonlinear dynamical system (DS) that generated observed time series. Recurrent Neural Networks (RNNs) are, in principle, powerful enough to approximate any underlying DS, but in their vanilla form suffer from the exploding vs. vanishing gradients pro…
Recent works have highlighted the strength of the Transformer architecture on sequence tasks while, at the same time, neural architecture search (NAS) has begun to outperform human-designed models. Our goal is to apply NAS to search for a better alternative to the Transformer. We first construct a large search space in…