This work reduces model size by 86.11% for recommender systems using 4-bit quantization.
problem Large memory consumption in embedding vectors for recommender systems.
method Post-training 4-bit quantization on embedding tables, including row-wise uniform quantization and codebook-based quantization.
result Consistently reduces accuracy degradation while significantly reducing model size.
Unified finetuning of all quantization degrees of freedom achieves state-of-the-art 4-bit quantization.
problem Achieving high accuracy in quantized neural networks while maintaining speed and resource constraints.
method Quantization-aware finetuning (QFT) that jointly optimizes all quantization degrees of freedom.
result 4-bit weight quantization results on-par with state-of-the-art (SoTA) within PTQ constraints.
New method corrects quantization errors in LLMs using low-rank matrices.
problem Correcting quantization errors in large language models.
method Introducing low-rank weight matrices to correct quantized activations in LLMs.
result Reduces accuracy gap with original model by more than 50% using low-rank matrices.
FP6 quantization outperforms INT4 in diverse generative tasks for LLMs.
problem Limited performance of 4-bit quantization methods in diverse generative tasks.
method Proposes a 4+2 design for FP6 quantization to achieve similar latency to INT4.
result FP6 quantization outperforms INT4 in various generative tasks, including code generation and summarization.
DFRot improves LLMs by reducing outlier and massive activation effects.
problem Reducing outlier and massive activation effects in rotated LLMs.
method Weighted loss function and orthogonal Procrustes transforms for rotation matrix refinement.
result DFRot achieves dual free (Outlier-Free and Massive Activation-Free) with significant improvements in perplexity.
Improved neural quantization reduces accuracy loss to less than 1% with 4-bit weights.
problem Reducing accuracy loss in neural quantization below 8-bits.
method Layer-wise calibration and integer programming to optimize bit-width allocation.
result Less than 1% accuracy degradation with 4-bit weights and activations.
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.
Deep networks run with low precision operations at inference time offer power and space advantages over high precision alternatives, but need to overcome the challenge of maintaining high accuracy as precision decreases. Here, we present a method for training such networks, Learned Step Size Quantization, that achieves…
Differentiable model compression adds noise to parameters during training.
problem Model compression for efficient deployment.
method Adding pseudo quantization noise to model parameters during training.
result Compresses model by more than 8x on ImageNet with 0.3% accuracy loss.
Model compression has gained a lot of attention due to its ability to reduce hardware resource requirements significantly while maintaining accuracy of DNNs. Model compression is especially useful for memory-intensive recurrent neural networks because smaller memory footprint is crucial not only for reducing storage re…
AdaRound improves post-training quantization of neural networks.
problem Improving the accuracy of quantized weights in neural networks.
method Adaptive rounding mechanism that adapts to data and task loss.
result AdaRound outperforms rounding-to-nearest and achieves state-of-the-art performance.
New method improves accuracy of quantized neural networks.
problem Accuracy drop in quantized neural networks, especially MobileNet family.
method Weight equalizing shift scaler, binary shifting to recover output range.
result Top-1 accuracy improved from 0.1% to 69.78% ~ 70.96% in MobileNets.
Post-training quantization saves resources for neural networks.
problem Implementing neural networks in resource-constrained hardware.
method Generalized post-training quantization method (GPFQ) with modifications for sparsity and error analysis.
result Error decays linearly with over-parametrization, showing minor loss of accuracy.
Recent machine learning methods use increasingly large deep neural networks to achieve state of the art results in various tasks. The gains in performance come at the cost of a substantial increase in computation and storage requirements. This makes real-time implementations on limited resources hardware a challenging …
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…
LSQ+ improves quantization of neural nets with Swish activations, achieving state-of-the-art results.
problem Quantization of neural nets with Swish activations, especially negative activations, leads to significant performance loss.
method Introduces learnable scale and offset parameters for asymmetric quantization, and uses MSE-based initialization for quantization parameters.
result Significantly outperforms LSQ for low-bit quantization of neural nets with Swish activations, achieving up to 5.6% gain with W2A2 quantization of EfficientNet-B0.
Model quantization is leveraged to reduce the memory consumption and the computation time of deep neural networks. This is achieved by representing weights and activations with a lower bit resolution when compared to their high precision floating point counterparts. The suitable level of quantization is directly relate…
Paper improves DNN accelerator robustness against bit errors with energy savings.
problem Bit errors in quantized DNN weights reduce energy efficiency.
method Combines robust fixed-point quantization, weight clipping, and random bit error training.
result Significantly improves robustness against random bit errors with high energy savings.
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…
OverQ increases model accuracy by handling outliers in neural networks with minimal hardware changes.
problem Handling outliers in neural network weights and activations for low-precision quantization.
method Overwrite quantization (OverQ) that opportunistically increases bitwidth for activation outliers.
result OverQ can handle over 90% of outliers and achieve +5% ImageNet Top-1 accuracy on a quantized ResNet-50 at 4 bits.
The paper analyzes how quantization affects the Fisher Information Matrix's dominant eigenvalue.
problem The impact of quantization on the Fisher Information Matrix's dominant eigenvalue.
method The study examines spectral perturbation of the empirical Fisher Information Matrix under in-distribution input and quantized parameter perturbations.
result A bound on the eigenvalue under quantization noise, showing it strictly exceeds the unperturbed value at leading order.
Low-bit training framework reduces energy consumption in CNNs.
problem Reducing energy consumption in convolutional neural networks.
method Low-bit training framework using MLS tensor format with dynamic quantization.
result Achieves superior trade-off between accuracy and bit-width.
We explore a new research direction in Bayesian variational inference with discrete latent variable priors where we exploit Kronecker matrix algebra for efficient and exact computations of the evidence lower bound (ELBO). The proposed "DIRECT" approach has several advantages over its predecessors; (i) it can exactly co…
Implementing large-scale deep neural networks with high computational complexity on low-cost IoT devices may inevitably be constrained by limited computation resource, making the devices hard to respond in real-time. This disjunction makes the state-of-art deep learning algorithms, i.e. CNN (Convolutional Neural Networ…
A new generation of manycore processors is on the rise that offers dozens and more cores on a chip and, in a sense, fuses host processor and accelerator. In this paper we target the efficient training of generalized linear models on these machines. We propose a novel approach for achieving parallelism which we call Het…
A hybrid neural network optimizes AI deployment on edge and cloud for energy efficiency.
problem Energy and resource constraints in edge devices for deep learning models.
method Conditionally deep hybrid neural network with quantized layers at edge and full-precision layers at cloud.
result Early classification at the edge reduces energy consumption by 5.5x on CIFAR-10 dataset.
Improves matrix multiplication throughput for asymmetric bit-width operands.
problem Matrix multiplications between asymmetric bit-width operands, especially 8- and 4-bit, are not efficiently handled by existing SIMD instructions.
method Proposes a new SIMD matrix multiplication instruction that uses mixed precision on inputs (8- and 4-bit) and accumulates into 16-bit output, improving throughput.
result Offers 2x improvement in throughput compared to existing symmetric-operand-size instructions, with negligible overflow.
Survey on quantization methods on Kähler manifolds.
problem None explicitly stated; focuses on methods.
method Deformation quantization, geometric quantization, Berezin-Toeplitz quantization, BV quantization.
result New relationships among quantization methods on Kähler manifolds.
The paper classifies quantizable functions and explores symmetry in quantization methods.
problem Classifying quantizable functions and understanding symmetry in quantization methods.
method Deformation quantization and geometric quantization methods are compared and classified.
result Formal quantizable functions are of a specific form and relate to Hamiltonian Killing vector fields.
This paper introduces a differentiable, scalable quantization method for neural networks.
problem Previous quantization methods lacked differentiability and scalability.
method The approach is differentiable and scalable, using bit-shifting and logarithmic quantization.
result The method achieves comparable accuracy to state-of-the-art approaches with less training time and lower inference cost.
StatQAT optimizes quantization for deep networks, reducing computational cost and memory usage.
problem Optimal quantization parameters selection for deep neural networks with diverse data distributions.
method Statistical error analysis framework for uniform and floating-point quantization, iterative and analytic quantizers designed for arbitrary and Gaussian-like distributions.
result Improved accuracy and stability in training low-precision neural networks.
Smart Quantization adapts binary and ternary quantization for neural networks.
problem Resource constraints in deploying neural networks on devices with limited resources.
method Adaptive combination of binary and ternary quantization with a regularization function.
result Adapts quantization depth during training to maintain high model accuracy.
This study optimizes quantized neural networks by considering model architecture and quantization types.
problem Optimizing quantized neural networks for low-power, high-throughput applications.
method Holistic approach including training methods and quantization-friendly architecture design.
result Deeper models are more sensitive to activation quantization, while wider models improve resilience to both weight and activation quantization.
Extends ONNX for quantized neural networks with new formats and operators.
problem Handling arbitrary-precision quantization in neural networks.
method Introduces new formats and operators in ONNX to represent quantized neural networks.
result Enabled representation of uniform quantization in neural networks.
New method for quantizing symplectic manifolds with Lagrangian bundles.
problem Quantization of symplectic manifolds with Lagrangian bundles.
method A new construction of strict deformation quantization.
result Established a correspondence between differential operators and principal symbols.
HMQ improves quantization for edge devices with mixed precision.
problem Efficient quantization for edge devices with uniform, power-of-two thresholds.
method Introduces HMQ, a mixed precision quantization block that repurposes Gumbel-Softmax for searching over quantization schemes.
result Achieves competitive and state-of-the-art results on ImageNet despite restrictions.
Introduces sheaf quantization, a topological approach to geometric quantization.
problem Topological realization of WKB-states in geometric quantization.
method Enhancement of constructible sheaves, Betti counterpart of Fukaya--Floer theory.
result Introduction to sheaf quantization as a topological realization of WKB-states.
Network quantization is an effective solution to compress deep neural networks for practical usage. Existing network quantization methods cannot sufficiently exploit the depth information to generate low-bit compressed network. In this paper, we propose two novel network quantization approaches, single-level network qu…
The article defines and compares two types of quantizations on compact manifolds.
problem Quantization on arbitrary compact smooth manifolds.
method Embedding into CP^n and inducing quantizations from there.
result Generalizations of earlier quantization methods.
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…
Image steganography is a procedure for hiding messages inside pictures. While other techniques such as cryptography aim to prevent adversaries from reading the secret message, steganography aims to hide the presence of the message itself. In this paper, we propose a novel technique for hiding arbitrary binary data in i…
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.
Proposes a robust neural network quantization method.
problem Training model's dependency on specific quantization methods.
method Intrinsic robustness to various quantization processes.
result Single model capable of operating at various bit-widths and policies.
Quantizes neural networks using frame theory for improved accuracy.
problem Improving neural network efficiency and accuracy through quantization.
method Sigma-Delta (ΣΔ) quantization with finite unit-norm tight frames. result Error bound between original and quantized neural networks derived.
The paper studies quantization on symplectic manifolds with real polarizations, comparing different quantization methods.
problem Quantization on compact symplectic manifolds with real polarizations.
method Geometric quantization, Toeplitz operators, Fourier transforms, asymptotic expansion of traces.
result Deformation quantization is realized through asymptotic traces of Toeplitz operators.
Geometric quantization of a Poisson manifold need not imply quantization of its symplectic leaves. We provide the leafwise geometric quantization of a Poisson manifold, seen as a foliated one, whose quantum algebra restricted to each leaf is quantized.
Meta learning optimizes neural network quantization for efficient inference.
problem Uniform bitwidth quantization is sub-optimal for neural network compression.
method Meta learning to automatically generate hybrid quantization policies.
result Meta learning outperforms uniform quantization and RL approaches.
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.