SOTAVerified

Model Compression

Model Compression is an actively pursued area of research over the last few years with the goal of deploying state-of-the-art deep networks in low-power and resource limited devices without significant drop in accuracy. Parameter pruning, low-rank factorization and weight quantization are some of the proposed methods to compress the size of deep networks.

Source: KD-MRI: A knowledge distillation framework for image reconstruction and image restoration in MRI workflow

Papers

Recently AddedMost HypedMost ActiveNeeds VerificationMost Verified

Showing 51–100 of 1356 papers

TitleStatusHype
Streamlining Redundant Layers to Compress Large Language ModelsCode1
FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix ApproximationCode1
Communication-Computation Trade-Off in Resource-Constrained Edge InferenceCode1
Forget the Data and Fine-Tuning! Just Fold the Network to CompressCode1
Computation-Efficient Knowledge Distillation via Uncertainty-Aware MixupCode1
CompRess: Self-Supervised Learning by Compressing RepresentationsCode1
Consistent Quantity-Quality Control across Scenes for Deployment-Aware Gaussian SplattingCode1
Generative Model-based Feature Knowledge Distillation for Action RecognitionCode1
Enhancing Cross-Tokenizer Knowledge Distillation with Contextual Dynamical MappingCode1
AD-KD: Attribution-Driven Knowledge Distillation for Language Model CompressionCode1
ADMM-NN: An Algorithm-Hardware Co-Design Framework of DNNs Using Alternating Direction Method of MultipliersCode1
Contrastive Distillation on Intermediate Representations for Language Model CompressionCode1
Enabling Lightweight Fine-tuning for Pre-trained Language Model Compression based on Matrix Product OperatorsCode1
Efficient and Robust Quantization-aware Training via Adaptive Coreset SelectionCode1
Towards Compact Neural Networks via End-to-End Training: A Bayesian Tensor Approach with Automatic Rank DeterminationCode1
Environmental Sound Classification on the Edge: A Pipeline for Deep Acoustic Networks on Extremely Resource-Constrained DevicesCode1
EarlyBERT: Efficient BERT Training via Early-bird Lottery TicketsCode1
Dynamic Slimmable NetworkCode1
Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and BetterCode1
Dual Relation Knowledge Distillation for Object DetectionCode1
Bidirectional Distillation for Top-K Recommender SystemCode1
Dynamic Channel Pruning: Feature Boosting and SuppressionCode1
Efficient On-Device Session-Based RecommendationCode1
EvoPress: Towards Optimal Dynamic Model Compression via Evolutionary SearchCode1
Backdoor Attacks on Federated Learning with Lottery Ticket HypothesisCode1
Aligned Structured Sparsity Learning for Efficient Image Super-ResolutionCode1
Distilling Linguistic Context for Language Model CompressionCode1
Basic Binary Convolution Unit for Binarized Image Restoration NetworkCode1
A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution RobustnessCode1
BERT-EMD: Many-to-Many Layer Mapping for BERT Compression with Earth Mover's DistanceCode1
Communication-Efficient Federated Learning through Adaptive Weight Clustering and Server-Side DistillationCode1
BERT-of-Theseus: Compressing BERT by Progressive Module ReplacingCode1
Distilled Split Deep Neural Networks for Edge-Assisted Real-Time SystemsCode1
Dynamic DNNs and Runtime Management for Efficient Inference on Mobile/Embedded DevicesCode1
Activation-Informed Merging of Large Language ModelsCode1
Bit-mask Robust Contrastive Knowledge Distillation for Unsupervised Semantic HashingCode1
CHEX: CHannel EXploration for CNN Model CompressionCode1
Communication-Efficient Diffusion Strategy for Performance Improvement of Federated Learning with Non-IID DataCode1
Distilling Object Detectors with Feature RichnessCode1
Learning Efficient Vision Transformers via Fine-Grained Manifold DistillationCode1
Clustered Sampling: Low-Variance and Improved Representativity for Clients Selection in Federated LearningCode1
Class Attention Transfer Based Knowledge DistillationCode1
Model LEGO: Creating Models Like Disassembling and Assembling Building BlocksCode1
CoA: Towards Real Image Dehazing via Compression-and-AdaptationCode1
Compacting, Picking and Growing for Unforgetting Continual LearningCode1
Discrimination-aware Channel Pruning for Deep Neural NetworksCode1
A Unified Pruning Framework for Vision TransformersCode1
Compression-Aware Video Super-ResolutionCode1
Comprehensive Knowledge Distillation with Causal InterventionCode1
Discrimination-aware Network Pruning for Deep Model CompressionCode1
Show:102550
← PrevPage 2 of 28Next →
All datasetsImageNetQNLI

Benchmark Results

#ModelMetricClaimedVerifiedStatus
1ADLIK-MO-ResNet50+W4A4Top-177.88—Unverified
2ADLIK-MO-ResNet50+W3A4Top-177.34—Unverified
3ResNet-18 + 4bit-1dim model compression using DKMTop-170.52—Unverified
4MobileNet-v1 + 4bit-1dim model compression using DKMTop-169.63—Unverified
5ResNet-18 + 2bit-1dim model compression using DKMTop-168.63—Unverified
6MobileNet-v1 + 2bit-1dim model compression using DKMTop-167.62—Unverified
7ResNet-18 + 4bit-4dim model compression using DKMTop-166.1—Unverified
8ResNet-18 + 2bit-2dim model compression using DKMTop-164.7—Unverified
9MobileNet-v1 + 4bit-4dim model compression using DKMTop-161.4—Unverified
10ResNet-18 + 1bit-1dim model compression using DKMTop-159.7—Unverified
#ModelMetricClaimedVerifiedStatus
1MobileBERT + 2bit-1dim model compression using DKMAccuracy82.13—Unverified
2MobileBERT + 1bit-1dim model compression using DKMAccuracy63.17—Unverified