A new GGN method speeds up training of deep neural networks for regression tasks.
problem Training deep neural networks efficiently for regression problems.
method Proposes a Gram-Gauss-Newton (GGN) algorithm for overparameterized neural networks.
result For sufficiently wide neural networks, GGN achieves quadratic convergence rate.
Negative step sizes improve second-order methods for neural networks.
problem Second-order methods discard negative curvature, limiting their effectiveness.
method Introduce negative step sizes in second-order methods combined with Wolfe line search.
result Negative step sizes lead to global convergence and improved performance.
Adaptive norms improve neural network optimization with second-order methods.
problem Optimizing neural networks with second-order methods.
method Regularized Newton methods with adaptive norms, interpreting adaptive gradient methods as trust region methods.
result Ellipsoidal constraints outperform spherical constraints in terms of backpropagations and asymptotic loss value.
A new method for optimizing deep neural networks using TKFAC.
problem Optimizing deep neural networks with second-order methods.
method Proposes Trace-restricted Kronecker-factored Approximate Curvature (TKFAC) for Fisher information matrix approximation.
result TKFAC improves performance on deep network architectures compared to state-of-the-art algorithms.
Second-order optimization speeds up deep hedging for complex options.
problem Hedging exotic options with market frictions in realistic markets.
method Second-order optimization scheme leveraging pathwise differentiability and Kronecker-factoring.
result Our method optimizes the policy in 1/4 the steps of standard optimization.
ODE2VAE learns latent dynamics for sequential data.
problem Learning latent dynamics for high-dimensional sequential data.
method Deep generative second order ODE model with Bayesian neural networks.
result State-of-the-art performance in long-term motion prediction and imputation.
New pruning method captures global correlations for efficient neural network inference.
problem Efficiently pruning neural networks for faster inference and reduced memory usage.
method Second-order structured pruning (SOSP-H) with innovative saliency-based approaches.
result SOSP-H scales to large-scale vision tasks and improves accuracy without compromising efficiency.
This paper uses second-order Poincaré inequalities to establish quantitative central limit theorems for Gaussian neural networks.
problem Establishing quantitative central limit theorems for Gaussian neural networks.
method Using second-order Poincaré inequalities to reduce the problem to computing the gradient and Hessian of the NN's output.
result Suboptimal rates of convergence for the NN's output due to the use of second-order Poincaré inequalities.
SOAR improves deep networks' robustness against adversarial examples.
problem Improving deep neural networks' robustness against adversarial examples.
method Formulated adversarial robustness problem under robust optimization framework, approximated loss function using second-order Taylor series expansion.
result SOAR significantly improves robustness of networks against adversarial perturbations.
New MPNNs match 2-WL, faster distinguishing graphs.
problem Improving graph neural network expressiveness.
method Introducing ℓ-walk MPNNs and second-order GNNs. result Walk MPNNs match 2-WL and can distinguish graphs faster.
Paper proposes a new method for efficient second-order neural network training.
problem Infeasibility of Hessian calculation and noisy second-order information in deep learning.
method Adopting complex-step directional derivative (CSFD) for accurate Hessian computation and designing an effective Newton Krylov procedure.
result Our method outperforms existing methods and often converges one-order faster.
Proposes a method to measure model parameter similarity for visual tasks.
problem Estimating relations between different visual tasks.
method LPS method using a second-order neural network to align model parameters and learn second-order similarity.
result Extensive experiments validate the effectiveness of the proposed method.
New models improve language generation by sharing intermediate states.
problem Language models struggle with past mistakes in sequence generation.
method Integrate second-order terms in hidden-state update, sharing intermediate states.
result Shared parametrization improves language modeling performance.
Efficient neural models for complex multi-hop reasoning tasks.
problem Complex multi-hop reasoning tasks in large knowledge bases.
method Differentiable neural models using symbolic knowledge bases, with a new operation for multi-hop template construction.
result Simple neural models achieve competitive performance on multi-hop reasoning tasks.
GradVis visualizes and analyzes deep neural network optimization surfaces efficiently.
problem Understanding and optimizing deep neural network training landscapes.
method Developed an open-source library GradVis for efficient visualization and analysis of optimization surfaces.
result GradVis enables plotting of optimization surfaces and trajectories for large networks.
Derives formulae for general permutation equivariant layers and presents a second order graph variational encoder.
problem Tackles the limitation of previous equivariant neural networks by considering permutations of matrices.
method Derives formulae for general permutation equivariant layers, including matrix permutations. Presents a second order graph variational encoder.
result Latent distribution of equivariant generative models must be exchangeable.
New algorithm trains neural networks in near-linear time, overcoming slow convergence issues.
problem Slow convergence and computational overhead in training deep neural networks.
method Reformulates Gauss-Newton iteration as an ℓ2-regression problem and uses Fast-JL dimension reduction.
result Achieves an O(mn)-time algorithm for training ReLU networks, near-linear in dimension.
SOC-ICNN expands neural network representational capacity by using conic optimization.
problem Restrictive representational capacity of ReLU-based ICNNs.
method Proposes SOC-ICNN architecture that uses Second-Order Cone Programming.
result SOC-ICNN strictly expands representational space without increasing complexity.
We define a second-order neural network stochastic gradient training algorithm whose block-diagonal structure effectively amounts to normalizing the unit activations. Investigating why this algorithm lacks in robustness then reveals two interesting insights. The first insight suggests a new way to scale the stepsizes, …
New insights on pruning deep networks by preserving function locality.
problem Designing effective pruning methods for deep neural networks.
method Revisited loss modeling using first and second order Taylor expansions, emphasizing locality.
result Both first and second order Taylor expansions can achieve similar performance in pruning.
Exact second-order optimization for deep learning reduces computational cost and improves performance.
problem Inadequate use of second-order optimization methods in deep learning due to high computational cost and non-convexity.
method Developed an exact stochastic second-order Newton method that addresses the non-convexity issue and provides an expression for the stochastic Hessian.
result Exact second-order Newton direction formula and its application in deep learning datasets.
WoodFisher improves neural network compression efficiency and accuracy.
problem Efficiently estimating inverse Hessian for neural network optimization.
method WoodFisher: a method to compute a faithful and efficient estimate of the inverse Hessian.
result WoodFisher significantly outperforms state-of-the-art methods for pruning neural networks.
SOLBP extends efficient inference to uncertain Bayesian networks.
problem Inference in uncertain Bayesian networks with second-order probabilities.
method Extends Loopy Belief Propagation to second-order Bayesian networks.
result Generates inferences consistent with sum-product networks, more efficient and scalable.
Deep neural nets approximate high-dimensional HJB equations efficiently.
problem Approximating solutions to high-dimensional HJB equations.
method Deep neural networks for approximating solutions.
result Deep neural networks can approximate solutions without the curse of dimensionality.
Paper uses deep learning to solve PDEs without supervision.
problem Solving elliptic PDEs without labeled data.
method Uses deep neural networks and least-squares functionals.
result Demonstrates effectiveness on 1D second-order elliptic PDEs.
New algorithm TURTLE outperforms MAML and meta-learner LSTM.
problem Learning new tasks quickly with limited data and compute power.
method Combining insights from MAML and meta-learner LSTM, TURTLE uses second-order gradients.
result TURTLE outperforms MAML and meta-learner LSTM in few-shot learning tasks.
Proposes a new stochastic quasi-Newton method with Nesterov's acceleration.
problem Improving convergence in large-scale non-convex optimization problems.
method Stochastic quasi-Newton method with Nesterov's accelerated gradient.
result Improved performance compared to classical and popular methods.
This paper proposes a family of online second order methods for possibly non-convex stochastic optimizations based on the theory of preconditioned stochastic gradient descent (PSGD), which can be regarded as an enhance stochastic Newton method with the ability to handle gradient noise and non-convexity simultaneously. …
NN-Turb generates turbulent velocity statistics using neural networks.
problem Creating a 1D field with turbulent velocity statistics.
method Fully-convolutional neural network (NN-Turb) to generate the field.
result NN-Turb generates a 1D field that satisfies Kolmogorov's 2/3 and 4/5 laws, exhibiting intermittency.
A new method for growing neural networks by splitting neurons, improving efficiency.
problem Optimizing neural network structures, especially for lightweight architectures.
method A progressive training approach using steepest descent to adaptively grow and split neurons.
result The method provides a computationally efficient way to optimize neural network structures.
The fundamental learning theory behind neural networks remains largely open. What classes of functions can neural networks actually learn? Why doesn't the trained network overfit when it is overparameterized? In this work, we prove that overparameterized neural networks can learn some notable concept classes, including…
DEO uses gradient information to escape saddle points in neural networks.
problem Training deep neural networks struggles with flat regions and saddle points.
method Dimer-Enhanced Optimization (DEO) uses gradient information to estimate curvature and escape saddle points.
result DEO improves training efficiency and performance compared to standard first-order methods.
Large-scale distributed training of deep neural networks suffer from the generalization gap caused by the increase in the effective mini-batch size. Previous approaches try to solve this problem by varying the learning rate and batch size over epochs and layers, or some ad hoc modification of the batch normalization. W…
A new optimisation method efficiently scales Hessian-vector products for neural networks.
problem Challenges in applying second-order quasi-Newton methods due to large Hessian and non-convexity.
method Proposes an optimisation algorithm that asymptotically uses the exact inverse Hessian with modified eigenvalues.
result Demonstrates scalability and comparable performance to other optimisation methods in neural networks.
Trust region and cubic regularization methods have demonstrated good performance in small scale non-convex optimization, showing the ability to escape from saddle points. Each iteration of these methods involves computation of gradient, Hessian and function value in order to obtain the search direction and adjust the r…
We propose a neural network-based algorithm for solving forward and inverse problems for partial differential equations in unsupervised fashion. The solution is approximated by a deep neural network which is the minimizer of a cost function, and satisfies the PDE, boundary conditions, and additional regularizations. Th…
Optimizes wide low-rank neural networks for reduced parameters and cost.
problem Reducing the number of learnable parameters in wide neural networks.
method Analyzed edge-of-chaos dynamics and derived formulae for optimal weight and bias variances.
result Optimal weight and bias variances for low-rank networks follow from multiplicative scaling.
Advanced optimization algorithms such as Newton method and AdaGrad benefit from second order derivative or second order statistics to achieve better descent directions and faster convergence rates. At their heart, such algorithms need to compute the inverse or inverse square root of a matrix whose size is quadratic of …
This article develops a statistical test for the null hypothesis of strict stationarity of a discrete time stochastic process in the frequency domain. When the null hypothesis is true, the second order cumulant spectrum is zero at all the discrete Fourier frequency pairs in the principal domain. The test uses a window …
The study derives generalization bounds for neural oscillators, improving their performance with regularization.
problem Quantifying the generalization capacities of neural oscillators.
method Using Rademacher complexity and squared Wasserstein-1 distances, the study derives theoretical upper PAC generalization bounds for neural oscillators.
result Theoretical bounds show polynomial growth in estimation errors with MLP size and time length, and regularization improves performance.
Deep neural nets solve complex stochastic control problems.
problem Solving stochastic optimal control problems with control multiplicative noise.
method Deep recurrent neural networks and LSTM.
result Deep learning algorithm solves complex stochastic control problems efficiently.
Neural operators solve families of 2BSDEs efficiently.
problem Solving infinite families of 2BSDEs on bounded domains.
method Introduces a mild generative neural operator model to approximate solutions.
result Solution operators can be approximated by neural operators with polynomial parameters.
SONODEs and ANODEs improve learning of second order dynamics.
problem Learning dynamics governed by second order laws.
method Extended adjoint sensitivity method and theoretical analysis of ANODEs.
result SONODEs and ANODEs can learn higher order dynamics efficiently.
Paper develops SINNOs for approximating stochastic processes.
problem Approximating stochastic processes with neural networks.
method Developed stochastic interpolation neural network operators (SINNOs) with random coefficients.
result Established boundedness, interpolation accuracy, and approximation capabilities of SINNOs.
A new neural network initialization method is proposed for faster and more accurate training.
problem Efficient initialization for training multi-layer feedforward neural networks.
method Initialization based on Stein's identity, using eigenvectors of cross-moment matrix.
result The SteinGLM method is faster and more accurate than other initialization methods.
Improved neural network ensembles using Stein Variational Newton updates.
problem Lack of efficient second-order information in current ensemble methods.
method Proposes a novel approximate Bayesian inference method integrating Stein Variational Newton updates with scalable Hessian approximations.
result Significantly faster convergence and more accurate posterior distribution approximations.
Study neural networks by mapping correlations, revealing essential statistics.
problem Understanding information processing in trained neural networks.
method Characterize neural network as distribution transformations, focusing on correlation functions.
result Higher-order correlations are crucial for internal layers, while input layer captures more.
We solve S-shaped utility portfolio selection with SD constraints using algorithms and neural networks.
problem Optimizing portfolios with S-shaped utility functions under SD constraints.
method First-order SD constraint solution, numerical algorithm for SSD, neural network approach.
result Effective numerical and neural network solutions for SSD constrained problems.