Quantized Adam reduces communication cost in deep learning training.
problem Reducing communication cost in distributed deep learning training.
method Gradient and weight quantization with error feedback in Adam.
result Proposed methods converge to first-order stationary points.
Optimal gradient quantization reduces communication costs in distributed deep learning.
problem High communication costs in distributed training of deep neural networks.
method Deduced optimal gradient quantization conditions for binary and multi-level quantization, developed novel schemes for dynamic quantization levels.
result Demonstrated superior performance of proposed quantization schemes on CIFAR and ImageNet datasets.
Paper develops a statistical framework for quantized training of deep neural networks.
problem Lack of theoretical understanding of gradient quantization in FQT.
method Presented a statistical framework for analyzing FQT algorithms, viewing quantized gradient as a stochastic estimator of QAT gradient.
result Developed two novel gradient quantizers with smaller variance than existing per-tensor quantizer.
A new method reduces communication in distributed learning by skipping less informative gradient updates.
problem Efficient communication in distributed machine learning.
method Quantizes and skips less informative gradients to reduce communication overhead.
result Proves linear convergence rate similar to gradient descent with significant communication savings.
To make deep neural networks feasible in resource-constrained environments (such as mobile devices), it is beneficial to quantize models by using low-precision weights. One common technique for quantizing neural networks is the straight-through gradient method, which enables back-propagation through the quantization ma…
BatchNorm helps train quantized networks by avoiding gradient explosion.
problem Training quantized neural networks is difficult due to gradient issues.
method Investigated the impact of BatchNorm on both full-precision and quantized networks.
result BatchNorm avoids gradient explosion in quantized networks, contrary to expectations.
We present Rotated Adaptive Tetra-iterated Quantizer (RATQ), a fixed-length quantizer for gradients in first order stochastic optimization. RATQ is easy to implement and involves only a Hadamard transform computation and adaptive uniform quantization with appropriately chosen dynamic ranges. For noisy gradients with al…
Differentially quantized gradient methods improve convergence in noisy communication channels.
problem Optimizing distributed learning with limited communication bandwidth and noise.
method Introduces Differential Quantization (DQ) to compensate for quantization errors in gradient descent.
result DQ-GD achieves the same contraction factor as unquantized GD at high bitrates, proving asymptotic optimality.
Study quantization effects on high-dimensional linear regression learning.
problem Understanding quantization's impact on learning high-dimensional linear regression models.
method Analyzes stochastic gradient descent for high-dimensional linear regression under various quantization targets.
result Establishes precise bounds on excess risk for different quantization schemes.
One of the most significant bottleneck in training large scale machine learning models on parameter server (PS) is the communication overhead, because it needs to frequently exchange the model gradients between the workers and servers during the training iterations. Gradient quantization has been proposed as an effecti…
Quantized Stochastic Primal-Dual Methods for Distributed Optimization
problem Distributed optimization with stochastic gradients and finite-bit communication
method q-PDGD, a quantized stochastic primal-dual method
result Linear contraction to an explicit neighborhood under RSI, O(1/k) convergence under PL inequality
Adaptive quantization improves SGD accuracy in data-parallel settings.
problem Fixed gradient quantization schemes lead to suboptimal performance in deep learning.
method Developed adaptive quantization schemes ALQ and AMQ that update compression schemes based on gradient statistics.
result Improved validation accuracy on CIFAR-10 and ImageNet datasets by 2% and 1% respectively.
Quantization-aware training can recover accuracy lost by post-training quantization.
problem Post-training quantization (PTQ) can fail sharply at aggressive bitwidths.
method A unified geometric framework that explains PTQ failure and QAT recovery.
result QAT has a useful bias that steers iterates back into the basin.
Data parallelism can boost the training speed of convolutional neural networks (CNN), but could suffer from significant communication costs caused by gradient aggregation. To alleviate this problem, several scalar quantization techniques have been developed to compress the gradients. But these techniques could perform …
A new gradient quantization scheme improves communication efficiency in distributed training.
problem Efficiently compressing gradients for parallel training of large models.
method Proposes a new gradient quantization scheme with theoretical guarantees and empirical performance.
result The new scheme matches and exceeds the performance of existing methods.
Quantized deep neural networks (QDNNs) are attractive due to their much lower memory storage and faster inference speed than their regular full precision counterparts. To maintain the same performance level especially at low bit-widths, QDNNs must be retrained. Their training involves piecewise constant activation func…
New insights into quantized neural networks reveal learning dynamics and generalization errors.
problem Understanding the impact of quantization hyperparameters on learning dynamics in high-dimensional models.
method Theoretical analysis and fixed-point analysis of STE dynamics in quantized models.
result STE training in quantized models converges to a plateau followed by a sharp drop in generalization error, influenced by quantization range.
Modern distributed training of machine learning models suffers from high communication overhead for synchronizing stochastic gradients and model parameters. In this paper, to reduce the communication complexity, we propose \emph{double quantization}, a general scheme for quantizing both model parameters and gradients. …
Quantized Variational Inference improves ELBO optimization with fast convergence.
problem Maximizing Evidence Lower Bound (ELBO) for variational inference.
method Optimal Voronoi Tesselation for variance-free gradients, Richardson extrapolation for asymptotic improvement.
result Quantized Variational Inference leads to fast convergence with comparable computational cost.
Bayesian Bits unifies quantization and pruning through gradient optimization.
problem Joint mixed precision quantization and pruning for efficient neural networks.
method Gradient-based optimization with a novel bit width decomposition and learnable stochastic gates.
result Bayesian Bits achieves better accuracy vs. efficiency trade-off compared to static bit width networks.
SMGD trains low-bit neural networks with memory constraints.
problem Training large neural networks with limited memory.
method Stochastic Markov Gradient Descent (SMGD).
result Encouraging numerical results and theoretical guarantees.
FrostNet improves INT8 quantization efficiency in mobile networks.
problem The importance of network architecture for optimal INT8 quantization.
method Quantization-aware training (QAT) with StatAssist and GradBoost, hardware-aware NAS.
result FrostNets achieve higher recognition accuracy with comparable latency when quantized.
We analyze the effect of quantizing weights and activations of neural networks on their loss and derive a simple regularization scheme that improves robustness against post-training quantization. By training quantization-ready networks, our approach enables storing a single set of weights that can be quantized on-deman…
New algorithm improves clustering and quantization using MMD.
problem Approximating probability distributions with weighted mixtures of Dirac measures.
method Gradient flow, mean shift, and MMD-optimal quantization.
result MSIP algorithm is more robust than state-of-the-art methods.
Quantized neural networks are vulnerable to adversarial attacks.
problem Adversarial robustness of quantized neural networks.
method Investigated adversarial robustness of quantized neural networks under different threat models.
result Quantization does not offer robust protection and results in gradient masking.
Mirror Descent framework for efficient neural network quantization.
problem Efficiently quantize large neural networks for resource-limited devices.
method Introduce Mirror Descent framework for constrained optimization of quantized neural networks.
result Mirror Descent framework achieves state-of-the-art performance in quantized neural networks.
Global-QSGD accelerates distributed training by up to 3.51%.
problem High communication overhead in distributed deep learning.
method Allreduce-compatible gradient quantization with theoretical guarantees.
result Global-QSGD accelerates distributed training by up to 3.51%.
vqSGD reduces communication in distributed optimization with convergence guarantees.
problem Reduction of communication cost in distributed optimization.
method Vector quantization schemes based on convex hull of a point set.
result Asymptotic reduction in communication cost with convergence guarantees.
AdaQuantFL reduces communication in federated learning by adaptively quantizing model updates.
problem Efficient communication of model updates in federated learning with high-dimensional models and limited bandwidth.
method AdaQuantFL uses adaptive quantization to reduce the number of bits for model updates while maintaining low error floor.
result AdaQuantFL converges in fewer communicated bits compared to fixed quantization levels, with minimal impact on accuracy.
Neural network quantization has become an important research area due to its great impact on deployment of large models on resource constrained devices. In order to train networks that can be effectively discretized without loss of performance, we introduce a differentiable quantization procedure. Differentiability can…
A new parametric method studies Willmore flows and energy quantization.
problem Understanding Willmore flows and their singularities.
method Parametric approach to Willmore gradient flows.
result For small-energy weak immersions, a unique solution exists.
We consider the problem of decentralized consensus optimization, where the sum of n smooth and strongly convex functions are minimized over n distributed agents that form a connected network. In particular, we consider the case that the communicated local decision variables among nodes are quantized in order to all…
Efficient deep neural network (DNN) inference on mobile or embedded devices typically involves quantization of the network parameters and activations. In particular, mixed precision networks achieve better performance than networks with homogeneous bitwidth for the same size constraint. Since choosing the optimal bitwi…
WaveQ uses sinusoidal regularization to optimize deep quantization for neural networks, improving both efficiency and accuracy.
problem Deep quantization reduces bitwidth but can lead to significant accuracy loss due to inter-layer dependencies.
method WaveQ employs sinusoidal regularization to learn multiple quantization parameters during gradient-based training, balancing compute efficiency and accuracy.
result WaveQ achieves accuracy preservation and efficiency gains across various deep networks, outperforming state-of-the-art techniques.
This paper tackles tensor recovery from noisy and multi-level quantized measurements.
problem Tensors from multi-level quantized measurements.
method Nonconvex optimization problem with alternating proximal gradient descent.
result The recovery error diminishes to zero with increasing tensor dimensions.
Quantized Neural Networks (QNNs) are often used to improve network efficiency during the inference phase, i.e. after the network has been trained. Extensive research in the field suggests many different quantization schemes. Still, the number of bits required, as well as the best quantization scheme, are yet unknown. O…
The Straight-Through Estimator (STE) is widely used for back-propagating gradients through the quantization function, but the STE technique lacks a complete theoretical understanding. We propose an alternative methodology called alpha-blending (AB), which quantizes neural networks to low-precision using stochastic grad…
We extend quantization-aware training to extreme model compression.
problem Maximizing model accuracy with minimal model size.
method Quantize a random subset of weights during training, allowing unbiased gradients through other weights.
result Established new state-of-the-art compromises between accuracy and model size.
In order to quantize the gate parameters of the LSTM (Long Short-Term Memory) neural network model with almost no recognition performance degraded, a new quantization method named Quantization Loss Re-Learn Method is proposed in this paper. The method does lossy quantization on gate parameters during training iteration…
BinaryConnect is generalized and proven to converge.
problem Understanding and improving neural network quantization.
method Refined analysis of BC, introduction of ProxConnect.
result ProxConnect achieves competitive performance.
As the size and complexity of models and datasets grow, so does the need for communication-efficient variants of stochastic gradient descent that can be deployed to perform parallel model training. One popular communication-compression method for data-parallel SGD is QSGD (Alistarh et al., 2017), which quantizes and en…
New method classifies manifold-valued data using Riemannian geometry.
problem Classifying data on curved Riemannian manifolds.
method Probabilistic Learning Vector Quantization on Symmetric Positive Definite Matrices.
result The method outperforms traditional Euclidean methods on manifold-valued data.
Training large machine learning models requires a distributed computing approach, with communication of the model updates being the bottleneck. For this reason, several methods based on the compression (e.g., sparsification and/or quantization) of updates were recently proposed, including QSGD (Alistarh et al., 2017), …
APoT quantization improves neural network efficiency and accuracy.
problem Efficiently quantizing weights and activations in neural networks.
method Constraining quantization levels as sums of Powers-of-Two terms, applying reparameterization, and weight normalization.
result 4-bit quantized ResNet-50 achieves 76.6% top-1 accuracy, 22% computational cost reduction.
New method quantizes neural networks for mobile devices.
problem High computational and memory costs of deep neural networks.
method Formulates quantization as a differentiable function.
result Quantization networks outperform state-of-the-art methods.
FleXOR trains fractional quantization for neural networks, improving accuracy and size.
problem Quantization limits to integer bits restricts compression and accuracy.
method Encryption algorithm with XOR gates for fractional bits during inference.
result FleXOR achieves high accuracy with fractional sub-1-bit weights.
Moniqua improves SGD convergence with quantized communication.
problem Efficiently communicating in decentralized SGD with limited bandwidth.
method Modulo quantized communication in decentralized SGD.
result Moniqua converges at the same rate as full-precision communication with less bits.
Due to its efficiency and ease to implement, stochastic gradient descent (SGD) has been widely used in machine learning. In particular, SGD is one of the most popular optimization methods for distributed learning. Recently, quantized SGD (QSGD), which adopts quantization to reduce the communication cost in SGD-based di…