DSSP improves deep learning training speed by dynamically adjusting staleness thresholds.
problem Time-consuming deep learning training on large datasets.
method Dynamic Stale Synchronous Parallel (DSSP) framework that adapts staleness threshold at runtime.
result DSSP converges faster and achieves higher accuracy than other paradigms.
Proposes DC-S3GD for efficient large-scale decentralized neural network training.
problem Training large-scale decentralized neural networks efficiently.
method Decentralized stale-synchronous version of DC-ASGD with gradient correction.
result Achieves state-of-the-art results in training Convolutional Neural Networks.
GA method reduces gradient staleness in cloud computing.
problem Gradient staleness in asynchronous SGD methods.
method Gap-Aware (GA) method that penalizes stale gradients linearly to the Gap.
result GA outperforms existing methods in final test accuracy.
Asynchronous SGD can speed up training with a trade-off of gradient staleness.
problem Asynchronous SGD suffers from gradient staleness, affecting convergence error.
method Theoretical analysis of the error-runtime trade-off considering random straggling delays.
result A method of gradually varying synchronicity in distributed SGD is proposed and demonstrated.
HPSGD speeds up DNN training by paralleling data sync with local training.
problem Low cluster utilization in distributed deep neural network training.
method Hierarchical Parallel SGD (HPSGD) with improved model updating for stale gradients.
result Significantly boosts distributed DNN training and reduces stale gradients.
This paper presents Rudra, a parameter server based distributed computing framework tuned for training large-scale deep neural networks. Using variants of the asynchronous stochastic gradient descent algorithm we study the impact of synchronization protocol, stale gradient updates, minibatch size, learning rates, and n…
Many distributed machine learning (ML) systems adopt the non-synchronous execution in order to alleviate the network communication bottleneck, resulting in stale parameters that do not reflect the latest updates. Despite much development in large-scale ML, the effects of staleness on learning are inconclusive as it is …
Distributed Stochastic Gradient Descent (SGD) when run in a synchronous manner, suffers from delays in waiting for the slowest learners (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect convergence. In this work we present a novel theoretical characteriz…
New asynchronous SGD algorithms achieve optimal performance in distributed learning.
problem Asynchronous training introduces staleness, complicating optimization analysis.
method Developed rigorous framework for asynchronous first order stochastic optimization.
result Asynchronous SGD can achieve optimal time complexity, matching synchronous methods.
Asynchronous federated modeling improves spatial data sharing without centralizing raw data.
problem Privacy and bandwidth constraints in distributed spatial data.
method Asynchronous federated modeling using low-rank Gaussian process approximations with block-wise optimization and adaptive strategies.
result Asynchronous federated modeling achieves synchronous performance and outperforms it in heterogeneous settings.
Unified proof for scalable personalized federated learning.
problem Personalized federated learning under asynchronous updates.
method Unified proof for asynchronous federated learning with bounded staleness applied to MAML and ME personalization frameworks.
result Unified proof for convergence to first-order stationary point for smooth and non-convex functions.
FedBuff improves federated learning scalability with asynchronous updates.
problem Limited scalability of federated learning with synchronous updates.
method Introduces asynchronous updates (staleness) in federated learning.
result Theoretical analysis shows improved convergence rate with boundedness removed.
Ringmaster LMO accelerates training in distributed systems by asynchronously updating neural networks.
problem Asynchronous training in distributed systems where workers compute gradients at different speeds.
method Introduces an asynchronous LMO-based momentum method for unconstrained stochastic nonconvex optimization.
result Establishes convergence guarantees and time complexity bounds for asynchronous LMO-based updates.
Stochastic gradient MCMC (SG-MCMC) has played an important role in large-scale Bayesian learning, with well-developed theoretical convergence properties. In such applications of SG-MCMC, it is becoming increasingly popular to employ distributed systems, where stochastic gradients are computed based on some outdated par…
DANA mitigates gradient staleness in asynchronous distributed SGD with momentum.
problem Gradient staleness in asynchronous distributed SGD with momentum.
method DANA: a novel technique for asynchronous distributed SGD with momentum that computes the gradient on an estimated future position of the model's parameters.
result DANA fully incorporates momentum in asynchronous training with almost no ramifications to final accuracy.
Sparsification improves convergence in asynchronous distributed SGD, even in the presence of staleness.
problem Staleness in asynchronous distributed SGD.
method Applied sparsification to reduce communication overheads in distributed asynchronous settings.
result The ergodic convergence rate of sparsified asynchronous SGD matches that of vanilla SGD, $\mathcal{O} \left( 1/\sqrt{T}
ight)$, even in the presence of staleness.
Asynchronous distributed stochastic gradient descent methods have trouble converging because of stale gradients. A gradient update sent to a parameter server by a client is stale if the parameters used to calculate that gradient have since been updated on the server. Approaches have been proposed to circumvent this pro…
Paper proposes DSP to accelerate deep learning model training by addressing locking and straggler problems.
problem Training deep neural networks is slow and inefficient due to locking and straggler issues.
method Layer-wise Staleness and DSP algorithm to handle locking and straggler problems.
result DSP achieves significant training speedup with stronger robustness than compared methods.
DG separates successes and failures by gating updates with advantage and surprisal.
problem Negative learning from surprising data in distributed reinforcement learning.
method DG gates each update with the product of advantage and surprisal, suppressing failures and preserving successes.
result DG outperforms other methods in various challenging reinforcement learning tasks.
Unified analysis of asynchronous-SGD algorithms for distributed learning.
problem Analyzing asynchronous-SGD in heterogeneous settings with varying speeds and data distributions.
method Unified convergence theory for non-convex smooth functions, including pure asynchronous SGD and its modifications.
result Unified convergence rates for various asynchronous algorithms, including novel methods.
Novel higher-order group synchronization for noisy local measurements on hypergraphs.
problem Synchronizing higher-order local measurements on hyperedges to global estimates on nodes.
method Message passing algorithm for global synchronization of higher-order measurements.
result Higher-order method outperforms standard pairwise synchronization methods in certain applications.
New method uses neural networks for accurate angle estimation in noisy conditions.
problem Accurately estimate angles from noisy measurements in various applications.
method Directed Graph Neural Networks (GNNSync) for end-to-end trainable framework.
result GNNSync achieves competitive performance, even at high noise levels.
Triangle fees adjust fees based on trade size and price movement, improving price accuracy and revenue.
problem Price staleness and low fee revenue in AMMs.
method Decreasing marginal fees proportional to price movement, creating incentives for price accuracy.
result Triangle fees strictly improve the Pareto frontier of price accuracy versus losses.
The sectoral synchronization observed for the Japanese business cycle in the Indices of Industrial Production data is an example of synchronization. The stability of this synchronization under a shock, e.g., fluctuation of supply or demand, is a matter of interest in physics and economics. We consider an economic syste…
Hybrid approach for large-scale network synchronization using KF and PTP.
problem Synchronization of large-scale networks in 5G.
method Combines Kalman Filtering and PTP for pairwise synchronization, and Factor Graphs and Belief Propagation for end-to-end synchronization.
result Error in offset estimation remains below 5 ns in simulations.
We analyze how an observer synchronizes to the internal state of a finite-state information source, using the epsilon-machine causal representation. Here, we treat the case of exact synchronization, when it is possible for the observer to synchronize completely after a finite number of observations. The more difficult …
New framework allows selective removal of stale data in option calibration.
problem Inability to remove old data from calibrated option pricing models without full retraining.
method Introduces operator-theoretic Gauss-Newton framework for selective forgetting.
result Provides stability guarantees and perturbation bounds for selective data removal.
New algorithm uses PSO to optimize DNN training parameters in distributed systems.
problem Reducing synchronization frequency in DNN training leads to poor convergence.
method Integrates PSO into distributed training to automatically compute new parameters.
result Proposed algorithm outperforms synchronous methods in distributed DNN training.
Techniques such as ensembling and distillation promise model quality improvements when paired with almost any base model. However, due to increased test-time cost (for ensembles) and increased complexity of the training pipeline (for distillation), these techniques are challenging to use in industrial settings. In this…
ShadowSync separates background synchronization for scalable distributed training.
problem Reducing synchronization overhead in distributed training for high scalability.
method Separates synchronization from training and runs it in the background.
result Achieves both high throughput and excellent model quality at scale.
New method synchronizes graphs with probability measures on rotations.
problem Synchronizing graphs with measure-valued edges over rotations.
method Formulated as maximization of cycle-consistency in probability measures over rotations, using Sinkhorn divergences.
result Proposes a nonparametric Riemannian particle optimization approach converging to global optimum under certain conditions.
Study predicts synchronization state of financial time series using cross-recurrence plots.
problem Predicting the state of synchronization of financial time series.
method Cross-correlation analysis and deep learning framework for predicting synchronization state based on cross-recurrence plots.
result Satisfactory performance in predicting synchronization state for certain pairs of stocks.
Efficiently estimates rotations with corrupted data.
problem Rotation synchronization under high corruption and noise.
method Message passing algorithm with reweighted least squares.
result Superior performance over state-of-the-art methods.
This paper proposes a general model for synchronized crowding behavior. An order parameter is introduced to quantify the level of synchronization which is shown a function of percentage of agents in reactive state. Further, synchronization is shown to be driven by the most active agents with the highest volatility. A t…
Study optimizes estimation of orthogonal and rotation matrices from noisy data.
problem Estimating orthogonal and rotation matrices from noisy data.
method Iterative polar decomposition algorithm initialized by spectral methods.
result Algorithm achieves optimal error rate of $(1+o(1)) rac{σ^2 d(d-1)}{2np}$.
This work addresses the instability in asynchronous data parallel optimization. It does so by introducing a novel distributed optimizer which is able to efficiently optimize a centralized model under communication constraints. The optimizer achieves this by pushing a normalized sequence of first-order gradients to a pa…
Solves complex clustering and rotation synchronization problem.
problem Challenges in classifying and synchronizing rotated objects into multiple categories.
method Semidefinite programming relaxations to solve the joint problem of community detection and synchronization.
result Exact recovery of community detection and synchronization when extending stochastic block model.
Spectral method for joint community detection and group synchronization.
problem Jointly detecting communities and synchronizing orthogonal groups in graphs.
method Spectral decomposition followed by CPQR factorization.
result Near-optimal guarantees for exact and stable recovery of cluster memberships and orthogonal transforms.
Spectral methods achieve near-optimal performance in orthogonal and permutation group synchronization.
problem Recovering group elements from pairwise measurements in computer vision.
method Spectral methods applied with the leave-one-out technique.
result Near-optimal performance bounds for orthogonal and permutation group synchronization established.
Study financial markets using synchronization measures and clustering algorithms.
problem Analyze high-frequency trading dynamics and market states.
method Ordinal pattern series, information-theoretic synchronization measure, clustering algorithms, Markov model.
result Identify two coherent seasons of centralized and decentralized synchronicity.
Networks of coupled dynamical systems provide a powerful way to model systems with enormously complex dynamics, such as the human brain. Control of synchronization in such networked systems has far reaching applications in many domains, including engineering and medicine. In this paper, we formulate the synchronization…
Machine learning predicts synchronization transitions in unknown systems.
problem Predicting synchronization transitions in systems with unknown equations.
method Developed a 'parameter-aware' machine learning scheme using reservoir computing or echo state networks.
result Machine learning accurately predicts synchronization transitions, including hysteresis loops.
New method linearizes nonlinear coupled oscillators on graphs.
problem Predicting global synchronization in nonlinear coupled oscillators on graphs.
method Latent dynamic filters learned through supervised matrix factorization.
result Latent dynamics filters enable effective prediction of global synchronization.
New approach predicts stock price synchronization using RNNs and LSTMs.
problem Forecasting synchronization of stock prices in the Indian market.
method Utilizing recurrence plots and CRQA for non-linear analysis, RNNs and LSTMs for prediction.
result Accuracy of 0.98 and F1 score of 0.83 in predicting stock price synchronization.
New method solves group synchronization with cycle-edge message passing.
problem Solving group synchronization with adversarial or uniform corruption and small noise.
method Cycle-edge message passing procedure using cycle consistency information.
result Exact recovery and linear convergence guarantees under adversarial corruption.
Adaptive synchronization improves deep reinforcement learning performance.
problem Fixed step size synchronization can cause loss of properly learned networks.
method Adaptive synchronization based on recent agent behavior.
result Adaptive synchronization leads to better performance in games.
A new algorithm removes stale observations in dynamic Bayesian optimization.
problem Optimizing functions that change over time, keeping track of the optimum.
method Wasserstein distance-based criterion to quantify relevancy, removing stale observations.
result W-DBO maintains good predictive performance and high sampling frequency.
We make policy optimization algorithms batch size-invariant by decoupling proximal and behavior policies.
problem Some policy optimization algorithms do not have batch size-invariance, leading to inefficiencies.
method We decouple the proximal policy from the behavior policy to achieve batch size-invariance.
result Our approach makes policy optimization algorithms more efficient and allows them to use stale data more effectively.