Pith. sign in

REVIEW 28 cited by

iDLG: Improved Deep Leakage from Gradients

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2001.02610 v1 pith:YXE2NXWC submitted 2020-01-08 cs.LG stat.ML

iDLG: Improved Deep Leakage from Gradients

classification cs.LG stat.ML
keywords gradientslabelsapproachdataground-truthextractlearningdeep
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

It is widely believed that sharing gradients will not leak private training data in distributed learning systems such as Collaborative Learning and Federated Learning, etc. Recently, Zhu et al. presented an approach which shows the possibility to obtain private training data from the publicly shared gradients. In their Deep Leakage from Gradient (DLG) method, they synthesize the dummy data and corresponding labels with the supervision of shared gradients. However, DLG has difficulty in convergence and discovering the ground-truth labels consistently. In this paper, we find that sharing gradients definitely leaks the ground-truth labels. We propose a simple but reliable approach to extract accurate data from the gradients. Particularly, our approach can certainly extract the ground-truth labels as opposed to DLG, hence we name it Improved DLG (iDLG). Our approach is valid for any differentiable model trained with cross-entropy loss over one-hot labels. We mathematically illustrate how our method can extract ground-truth labels from the gradients and empirically demonstrate the advantages over DLG.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 28 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. From Efficiency to Leakage -- Privacy Backdoor in Federated Language Model Fine-Tuning

    cs.CR 2026-06 conditional novelty 7.0

    NeuroImprint attack assigns isolated memorization neurons to training samples in PEFT adapters, enabling closed-form reconstruction of 59-79% of samples across BERT, GPT-2, Qwen2, and Llama3.2 on multiple datasets.

  2. TIGER: Inverting Transformer Gradients via Embedding-Subspace Distance Optimization

    cs.CR 2026-06 unverdicted novelty 7.0

    TIGER turns the low-rank attention gradient subspace into a differentiable objective for continuous embedding optimization, improving reconstruction quality and robustness over prior discrete token tests especially un...

  3. Loss Landscape Poisoning: Targeted Extraction of Unseen Training Data from LLMs

    cs.CR 2026-06 unverdicted novelty 7.0

    Poisoning training data reshapes the loss landscape to enable targeted extraction of unseen data from LLMs with high success rates in language and vision-language models.

  4. TriShield: Zero-Utility-Loss Defense Against Privacy Backdoors in Federated Language Model Fine-Tuning via Orthogonal Gradient Projection and Optimizer State Entanglement

    cs.LG 2026-07 conditional novelty 6.0

    TriShield combines artifact detection, Adam momentum pre-entanglement, and SVD task-subspace projection to drive NeuroImprint reconstruction to 0% with claimed near-zero utility loss.

  5. BettiSplit: Topology-Guided Privacy-Aware Split Learning Against Feature Inversion and Gradient Leakage

    cs.LG 2026-07 conditional novelty 6.0

    Persistent Betti-1 of smashed activations is proposed as an attack-free indicator of feature-inversion risk in split learning, used for split selection and regularization.

  6. FedCVESA: Taking Away Training Data in Federated Learning via Correlation Value Encoding and Segmented Aggregation

    cs.LG 2026-07 conditional novelty 6.0

    A malicious FL server can steal private training images by encoding them into model parameters via a correlation regularizer and preserving them through segmented aggregation.

  7. TL++: Accuracy and Privacy Preserving Traversal Learning for Distributed Intelligent Systems

    cs.LG 2026-06 unverdicted novelty 6.0

    TL++ recovers centralized mini-batch gradients via virtual batches in split learning and adds secret sharing for cut-layer tensors, achieving 91.41% accuracy on CIFAR-10 with 13x lower communication than full-model sync.

  8. Exposing the Illusion of Erasure in Knowledge Editing for LLMs

    cs.LG 2026-06 unverdicted novelty 6.0

    Knowledge editing methods redistribute and suppress rather than overwrite facts in LLMs, creating narrow vulnerable regions in representation space that adversarial prompts can exploit.

  9. HADES: Privacy-Preserving Federated Learning via Selective Feature Encryption and Hybrid Model Fusion

    cs.CR 2026-06 unverdicted novelty 6.0

    HADES selectively encrypts privacy-sensitive features identified by PCA in federated learning, trains hybrid encrypted and plaintext networks, and fuses them to match vanilla FL accuracy with reduced overhead and bett...

  10. Secure Aggregation with Top-K Sparsification in Decentralized Federated Learning

    cs.IT 2026-06 unverdicted novelty 6.0

    A sparse secure aggregation scheme for decentralized federated learning that supports top-K sparsification, user dropouts, and collusion while preserving model accuracy at 1% sparsity.

  11. Local Differential Privacy via Dynamic Quantization in Distributed Online Stochastic Optimization

    math.OC 2026-05 unverdicted novelty 6.0

    Dynamic quantization provides local differential privacy in distributed online stochastic optimization with almost sure convergence on directed graphs.

  12. Response-Conditioned Parallel-to-Sequential Orchestration for Multi-Agent Systems

    cs.CL 2026-05 unverdicted novelty 6.0

    Nexa learns a response-conditioned policy that starts with parallel agent execution and adds at most one round of sequential message passing via a predicted sparse DAG, strictly subsuming pure parallel mode.

  13. On What We Can Learn from Low-Resolution Data

    cs.LG 2026-05 unverdicted novelty 6.0

    Low-resolution data improves high-resolution model performance when high-resolution samples are limited, via KL-divergence bounds and experiments on vision transformers and CNNs.

  14. Federated User Behavior Modeling for Privacy-Preserving LLM Recommendation

    cs.IR 2026-04 unverdicted novelty 6.0

    SF-UBM enables privacy-preserving cross-domain LLM recommendation by federating semantic item representations, distilling domain knowledge, and aligning preferences into LLM soft prompts.

  15. Approximate and Weighted Data Reconstruction Attack in Federated Learning

    cs.LG 2023-08 unverdicted novelty 6.0

    An interpolation approximation plus layer-weighted loss enables stronger data reconstruction attacks on FedAvg federated learning, with experiments showing gains on image data over prior methods.

  16. Privacy Leakage in Federated Learning in Radiology Reports: A Comparative Evaluation of Tokenizer-Driven Privacy Risks

    cs.LG 2026-07 reject novelty 5.0

    Up to 44% of radiology report sentences were exactly reconstructed from federated-learning gradients in this worst-case attack, with the RadBERT tokenizer leaking the most—but the paper's own re-run did not reproduce ...

  17. Profiling Privacy Preservation Against Gradient Inversion Attacks in Tabular Federated Learning

    cs.LG 2026-05 unverdicted novelty 5.0

    Empirical study finds small client batches most vulnerable to gradient inversion in tabular FL, FT-Transformer harder to invert than MLP baselines, and aggregate accuracy metrics can overstate exact record recovery.

  18. FIRMA: FIbonacci Ring Model Aggregation for Privacy-preserving Federated Learning

    cs.LG 2026-05 unverdicted novelty 5.0

    FIRMA introduces Fibonacci ring aggregation protocols for server-free federated learning that maintain private heads and achieve higher accuracy than FedAvg under label skew across multiple benchmarks and heterogeneit...

  19. LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections

    cs.LG 2026-05 unverdicted novelty 5.0

    LightSplit uses non-invertible orthogonal projections as an information bottleneck in split learning to reduce transmitted dimensionality by 32x while retaining more than 95% accuracy and limiting reconstruction risk.

  20. UMEDA: Unified Multi-modal Efficient Data Fusion for Privacy-Preserving Graph Federated Learning via Spectral-Gated Attention and Diffusion-Based Operator Alignment

    cs.LG 2026-05 unverdicted novelty 5.0

    UMEDA is a new graph federated learning method that uses low-rank spectral filtering and diffusion over a shared integral operator to fuse multi-modal data privately, outperforming baselines on MM-Fi and RELI11D under...

  21. Scalable and Private Federated Learning Using Distributed Differential Privacy and Secure Aggregation

    cs.CR 2026-04 conditional novelty 5.0

    A blacklist/whitelist-guided prompt optimization plus diffusion pipeline produces de-identified chest X-rays that retain enough pathology for competitive report-generation training while cutting patient-identity class...

  22. Scalable and Private Federated Learning Using Distributed Differential Privacy and Secure Aggregation

    cs.CR 2026-04 unverdicted novelty 5.0

    DDP-SA combines client-side Laplace noise perturbation with full-threshold additive secret sharing to let federated learning servers reconstruct only aggregated noisy gradients without exposing individual client updates.

  23. FedSpy-LLM: Towards Scalable and Generalizable Data Reconstruction Attacks from Gradients on LLMs

    cs.CR 2026-04 unverdicted novelty 5.0

    FedSpy-LLM uses gradient decomposition and iterative alignment to reconstruct larger batches and longer sequences of training data from LLM gradients in federated settings, including with PEFT methods.

  24. Don't Trust the AI Ecosystem: Analyzing Privacy Leakage in Compromised Open-Source Components

    cs.CR 2026-07 conditional novelty 4.0

    An attacker who controls a training script can store exact copies of private images inside model weights and extract them losslessly after training.

  25. DPDL: Towards Differential Privacy Preservation in Decentralized Stochastic Learning on Non-IID Data

    cs.LG 2026-06 unverdicted novelty 4.0

    DPDL perturbs cross-gradients with Gaussian noise for DP and applies cosine similarity calibration to support linear speedup in decentralized training on non-IID data.

  26. SafeLM: Unified Privacy-Aware Optimization for Trustworthy Federated Large Language Models

    cs.CR 2026-04 unverdicted novelty 4.0

    SafeLM unifies privacy-preserving federated LLM training with Paillier encryption, attack defenses, contrastive grounding, and binarized aggregation to achieve 98% harmful content detection, 96.9% less communication, ...

  27. Code-Poisoning Property Inference Attacks

    cs.CR 2026-07 reject novelty 3.0

    Code-level access lets CPPIA copy dataset proportions into the model's architecture as a hardcoded backdoor, so the 'inference' is just decoding what the attacker already computed.

  28. Knowledge Distillation in Federated Learning: a Survey on Long Lasting Challenges and New Solutions

    cs.LG 2024-06 unverdicted novelty 2.0

    A survey organizing knowledge distillation techniques for addressing privacy, heterogeneity, communication, and personalization challenges in federated learning.