SGD converges with perturbed forward-backward passes, explained by geometric amplification.
problem Analyzing convergence of SGD with perturbed forward-backward passes in composite optimization.
method Characterized propagation and amplification of perturbations, derived convergence guarantees for non-convex and PL objectives.
result Perturbations cascade through the computational graph, affecting convergence order under specific conditions.
Single-pass method estimates neural network uncertainty.
problem Uncertainty estimation in deep learning requires multiple passes.
method Probabilistic reasoning over neural network depths.
result Single forward pass for uncertainty estimation.
SHINE uses forward pass quasi-Newton matrices to approximate Jacobian inverses for faster bi-level optimization.
problem Efficiently solving bi-level optimization problems with large Jacobian matrices.
method Proposes using quasi-Newton matrices from the forward pass to approximate the inverse Jacobian matrix.
result Empirically shows SHINE reduces computational cost of the backward pass for various problems.
Factor graphs have recently gained increasing attention as a unified framework for representing and constructing algorithms for signal processing, estimation, and control. One capability that does not seem to be well explored within the factor graph tool kit is the ability to handle deterministic nonlinear transformati…
PFP-BNNs offer a fast, deterministic approach to Bayesian neural networks.
problem Limited uncertainty handling in traditional neural networks restricts their use in safety-critical settings.
method Probabilistic Forward Pass (PFP) approximates Stochastic Variational Inference (SVI) for efficient BNNs.
result PFP-BNNs achieve up to 4200x speedup over SVI-BNNs while maintaining similar accuracy and uncertainty.
Paper introduces efficient uncertainty estimation in LLMs without multiple forward passes.
problem Accurate uncertainty quantification in LLMs remains challenging.
method Evidential Knowledge Distillation to create compact student models.
result Efficient uncertainty estimation achieved with single forward pass.
This paper introduces Selective-Backprop, a technique that accelerates the training of deep neural networks (DNNs) by prioritizing examples with high loss at each iteration. Selective-Backprop uses the output of a training example's forward pass to decide whether to use that example to compute gradients and update para…
DG improves policy gradient efficiency by selectively backpropagating only valuable samples.
problem Expensive backward passes in policy gradient methods reduce efficiency.
method Introduces 'delight' as a forward-pass signal of learning value and a Kondo gate to selectively backpropagate.
result Selective backpropagation reduces backward pass costs without sacrificing learning quality.
SDPA is shown to be an optimal transport problem in deep learning.
problem The mathematical foundation and optimization perspective of SDPA.
method SDPA is shown to be the exact solution to a degenerate, one-sided Entropic Optimal Transport (EOT) problem.
result The SDPA mechanism is a principled mechanism where the forward pass performs optimal inference and the backward pass implements a rational, manifold-aware learning update.
We reduce training time in convolutional networks (CNNs) with a method that, for some of the mini-batches: a) scales down the resolution of input images via downsampling, and b) reduces the forward pass operations via pooling on the convolution filters. Training is performed in an interleaved fashion; some batches unde…
LayerNorm transformers have dead directions that can be read from their parameters alone.
problem Locating dead directions in LayerNorm transformers
method Using the inverse-scale direction of LayerNorm affine parameters
result Predicted dead direction matches measured bottom singular direction
A method makes particle filters differentiable without altering their forward pass.
problem Compatibility issues between particle filters and automatic differentiation.
method Introduces a correction to particle weights using the stop-gradient operator.
result Automatic differentiation produces good estimators for gradients and second-order derivatives.
Neural networks offer high-accuracy solutions to a range of problems, but are costly to run in production systems because of computational and memory requirements during a forward pass. Given a trained network, we propose a techique called Deep Learning Approximation to build a faster network in a tiny fraction of the …
Proposes a deep model for Bayesian quantile regression without Gaussian assumptions.
problem Uncertainty quantification from single forward-pass models is computationally expensive and restrictive.
method Deep evidential learning for Bayesian quantile regression.
result Achieves calibrated uncertainties on non-Gaussian distributions.
Simple neural net outperforms complex uncertainty methods.
problem Reliable uncertainty estimation from deterministic models.
method A simple baseline using a single softmax neural net with residual connections and spectral normalization.
result Simple neural net outperforms DUQ and SNGP on uncertainty prediction.
Backpropagation-free RL method trains layers using local signals.
problem Vanishing or exploding gradients in backpropagation-based RL.
method Local pairwise distance matching for layer-wise training without backpropagation.
result Backpropagation-free method achieves competitive performance and stability.
A new model improves uncertainty estimation in deep learning.
problem Deep Kernel Learning (DKL) produces unreliable uncertainty estimates.
method Proposed a bi-Lipschitz constraint to preserve distances in feature space.
result DUE model outperforms previous DKL and other methods in uncertainty quality.
Convolutional neural networks (CNN) have led to many state-of-the-art results spanning through various fields. However, a clear and profound theoretical understanding of the forward pass, the core algorithm of CNN, is still lacking. In parallel, within the wide field of sparse approximation, Convolutional Sparse Coding…
We prove here a general closed-form expansion formula for forward-start options and the forward implied volatility smile in a large class of models, including the Heston stochastic volatility and time-changed exponential Lévy models. This expansion applies to both small and large maturities and is based solely on the p…
We propose a new variational inference algorithm for learning in Gaussian Process State-Space Models (GPSSMs). Our algorithm enables learning of unstable and partially observable systems, where previous algorithms fail. Our main algorithmic contribution is a novel approximate posterior that can be calculated efficientl…
MPLP learns neural network weights by treating operations as message-passing agents.
problem Training neural networks using gradient-based methods.
method MPLP abstracts neural network operations as message-passing agents, updating internal states and passing messages.
result MPLP outperforms traditional gradient-based methods on simple feed-forward neural networks.
As deep neural networks are increasingly being deployed in practice, their efficiency has become an important issue. While there are compression techniques for reducing the network's size, energy consumption and computational requirement, they only demonstrate empirically that there is no loss of accuracy, but lack for…
CPS solves inverse problems using forward passes and constrained particle seeking.
problem Solving inverse problems with limited forward observation information.
method Gradient-free approach that reformulates inverse problem as constrained optimization.
result CPS achieves results comparable to gradient-based methods while outperforming alternatives.
New method estimates data influence efficiently by leveraging test samples.
problem Efficiently estimating influence of training data on model predictions.
method Mirrored Influence Hypothesis, forward pass for test samples.
result Significant improvement in efficiency over existing methods.
Rich and complex time-series data, such as those generated from engineering systems, financial markets, videos or neural recordings, are now a common feature of modern data analysis. Explaining the phenomena underlying these diverse data sets requires flexible and accurate models. In this paper, we promote Gaussian pro…
Single deep model detects out-of-distribution data with single forward pass.
problem Detecting out-of-distribution data points in neural networks.
method Deterministic uncertainty quantification (DUQ) using gradient penalty for reliable detection.
result Single model outperforms or matches ensemble methods in out-of-distribution detection.
This note presents in a technical though hopefully pedagogical way the three most common forms of neural network architectures: Feedforward, Convolutional and Recurrent. For each network, their fundamental building blocks are detailed. The forward pass and the update rules for the backpropagation algorithm are then der…
Model trains passing events on a bridge using multilevel Gaussian process.
problem Represent aggregate train-passing events from a bridge monitoring system.
method Formulate a combined model with low-rank approximation hierarchical Gaussian process, incorporating domain expertise as constraints.
result Allow for simulation of previously unobserved train types.
Single model predicts multiple subnetworks for robustness.
problem Efficiently ensemble neural networks for robustness and uncertainty.
method Multi-input multi-output (MIMO) configuration to train independent subnetworks.
result Significant improvement in negative log-likelihood, accuracy, and calibration error.
We introduce graph normalizing flows: a new, reversible graph neural network model for prediction and generation. On supervised tasks, graph normalizing flows perform similarly to message passing neural networks, but at a significantly reduced memory footprint, allowing them to scale to larger graphs. In the unsupervis…
Google's Cloud TPUs are a promising new hardware architecture for machine learning workloads. They have powered many of Google's milestone machine learning achievements in recent years. Google has now made TPUs available for general use on their cloud platform and as of very recently has opened them up further to allow…
This paper introduces a new framework for data efficient and versatile learning. Specifically: 1) We develop ML-PIP, a general framework for Meta-Learning approximate Probabilistic Inference for Prediction. ML-PIP extends existing probabilistic interpretations of meta-learning to cover a broad class of methods. 2) We i…
DDEQs extend DEQs to discrete measure inputs using Wasserstein gradient flows.
problem Applying DEQs to discrete measure inputs like sets or point clouds.
method Wasserstein gradient flows for finding fixed points of discrete measures under permutation-invariance.
result DDEQs can compete with state-of-the-art models in tasks like point cloud classification and completion.
Paper introduces FoMoH for optimization without backpropagation.
problem Optimizing machine learning models without backpropagation.
method Second-order hyperplane search, forward-mode stochastic gradient method, hyper-dual numbers, FoMoH.
result Developed a novel optimization algorithm that avoids backpropagation.
Density-Regression improves deep uncertainty estimation with faster inference.
problem Efficient uncertainty estimation under distribution shifts with modern deep models.
method Leverages density function for fast inference and distance-aware feature space.
result Density-Regression achieves competitive uncertainty estimation performance.
OLS is a special case of Transformer, revealing its linear nature.
problem Understanding the statistical essence of Transformer architecture.
method Algebraic proof and spectral decomposition of covariance matrix.
result Attention mechanism in Transformers is mathematically equivalent to OLS.
The article constructs a forward utility for markets with multiple default risks.
problem Characterizing forward performance processes in a market with multiple default risks.
method Using Jacod-Pham decomposition and recursive BSDEs, the article constructs a forward utility and proves its existence and uniqueness.
result The article identifies the risk-sensitive long-run growth rate of the optimal wealth process in a stochastic factor model with ergodic dynamics.
FP uses random projections to train networks without feedback, achieving comparable performance to backpropagation.
problem Training neural networks without feedback from downstream layers.
method Forward Projection (FP) method that uses randomised nonlinear projections and closed-form regression.
result FP achieves comparable generalisation to backpropagation methods with a single forward pass, offering significant speedup.
Exact guidance for discrete data improves posterior sampling efficiency.
problem Inefficient guidance for discrete data in posterior sampling.
method Derive exact transition rate for desired distribution given learned discrete flow matching model.
result Significantly improved efficiency with single forward pass per sampling step.
A novel GNN architecture improves link prediction by combining positive and negative samples.
problem Link prediction accuracy limited by node-wise methods and increased model complexity of edge-wise methods.
method Proposes a novel GNN architecture that uses both positive and negative samples to inform node embeddings.
result The method retains the inference speed of node-wise models while achieving competitive accuracy with edge-wise alternatives.
New method reduces diffusion model function evaluations for discrete data.
problem High computational burden in generating samples from masked diffusion models.
method Modified causal attention mask and speculative sampling mechanism for non-factorized predictions.
result Achieved ~2x reduction in required network forward passes.
Dropout and RaM become equivalent in large ResNets as depth and width increase.
problem Improving performance in deep learning models.
method Comparing Dropout and Random Gradient Masking in ResNets.
result Dropout and RaM converge to the same large-scale limiting dynamics in ResNets.
Audit financial machine learning workflows to detect spurious predictability.
problem Spurious predictability in financial machine learning models.
method Falsification audit testing predictive workflows against synthetic environments.
result Many apparent financial predictions are artifacts, not genuine.
Backpropagation-free trunk training improves model performance on various benchmarks.
problem Memory inefficiency and noisy gradient estimates in deep network training.
method Split Forward Gradient (Split-FG) method that splits network into trunk and head, estimating only trunk gradient.
result Split-FG achieves better performance than pure forward-gradient training and backpropagation on various benchmarks.
DistPred provides a fast, distribution-free method for regression and forecasting.
problem Deterministic point estimates in regression and prediction tasks.
method Transforming proper scoring rules into a differentiable form and using it as a loss function.
result Achieved state-of-the-art performance and significantly improved computational efficiency.
Gated Linear Units (arXiv:1612.08083) consist of the component-wise product of two linear projections, one of which is first passed through a sigmoid function. Variations on GLU are possible, using different nonlinear (or even linear) functions in place of sigmoid. We test these variants in the feed-forward sublayers o…
Neural network training is computationally and memory intensive. Sparse training can reduce the burden on emerging hardware platforms designed to accelerate sparse computations, but it can affect network convergence. In this work, we propose a novel CNN training algorithm Sparse Weight Activation Training (SWAT). SWAT …
We propose Sideways, an approximate backpropagation scheme for training video models. In standard backpropagation, the gradients and activations at every computation step through the model are temporally synchronized. The forward activations need to be stored until the backward pass is executed, preventing inter-layer …