Tagged articles

adversarial training

17 articles · Page 1 of 1

Jul 19, 2026 · Artificial Intelligence

Alibaba Security AGI Unveils Three LLMs, 8B Model Beats GPT‑5.4 on Multiple Safety Metrics

Alibaba’s Security AGI lab introduced three Yuvion LLMs—8B, 32B, and a 32B Agent—trained on Qwen‑3, and demonstrated that the 8B model already surpasses most SOTA baselines while the 32B variants achieve top rankings in comprehensive safety, adversarial, and business‑level evaluations, outpacing GPT‑5.4 and Qwen‑3‑Max.

AI safetyAgentAlibaba

0 likes · 14 min read

Alibaba Security AGI Unveils Three LLMs, 8B Model Beats GPT‑5.4 on Multiple Safety Metrics

Bilibili Tech

Oct 31, 2025 · Artificial Intelligence

RIVAL: Adversarial RL Framework Elevates Conversational Subtitle Translation

RIVAL (Reinforcement Learning with Iterative and Adversarial Optimization) introduces an adversarial game between a reward model and a translation LLM, combining qualitative preference rewards with quantitative metrics like BLEU, to overcome distribution shift in RLHF and achieve superior performance on conversational subtitle and WMT translation tasks.

BLEULLMadversarial training

0 likes · 13 min read

RIVAL: Adversarial RL Framework Elevates Conversational Subtitle Translation

Data Party THU

Sep 22, 2025 · Artificial Intelligence

How to Secure Large‑Model Training: Practical Techniques and Real‑World Cases

This article systematically examines the major security challenges of large‑model training—including data leakage, adversarial attacks, bias, and supply‑chain risks—and presents concrete solutions such as differential privacy, federated learning, adversarial training, backdoor detection, and lifecycle protection to guide practitioners toward safer AI deployments.

AI safetyDifferential Privacyadversarial training

0 likes · 14 min read

How to Secure Large‑Model Training: Practical Techniques and Real‑World Cases

Tencent Cloud Developer

Mar 25, 2025 · Artificial Intelligence

Knowledge Distillation in Diffusion Models: Techniques and Applications

The article explains how knowledge distillation transfers capabilities from large to smaller diffusion models, covering hard and soft labels, temperature scaling, and contrasting it with data distillation, while detailing techniques such as consistency models, progressive distillation, adversarial distillation, and adversarial post‑training for model compression and step reduction.

adversarial post-trainingadversarial trainingconsistency models

0 likes · 19 min read

Knowledge Distillation in Diffusion Models: Techniques and Applications

AntTech

Jan 6, 2025 · Artificial Intelligence

2024 Security and Trusted AI Research Highlights from Alibaba, Tsinghua, Zhejiang, and Partner Institutions

This article presents sixteen peer‑reviewed research papers published in top conferences and journals in 2024, covering trusted AI, large‑model applications, network security, adversarial training, deep‑fake detection, secure inference, and related topics from collaborations among Alibaba, Tsinghua, Zhejiang, and other leading institutions.

AI securitySecure InferenceTrusted AI

0 likes · 27 min read

2024 Security and Trusted AI Research Highlights from Alibaba, Tsinghua, Zhejiang, and Partner Institutions

Ops Development & AI Practice

Jul 8, 2024 · Artificial Intelligence

Essential Denoising Techniques for Training Large AI Models

This article outlines key denoising methods—including data cleaning, augmentation, regularization, adversarial training, and self‑supervised learning—that improve the performance, generalization, and robustness of large neural network and transformer models.

Data AugmentationDenoisingSelf-supervised Learning

0 likes · 5 min read

Essential Denoising Techniques for Training Large AI Models

Baobao Algorithm Notes

Jun 3, 2024 · Artificial Intelligence

Can Adversarial Training Make Retrieval‑Augmented Generators More Robust?

Recent arXiv work introduces ATM, an adversarially‑tuned multi‑agent system that iteratively pits a fake‑knowledge attacker against a generator, dramatically improving retrieval‑augmented language models’ resistance to hallucinated content and boosting performance on knowledge‑intensive benchmarks, even with noisy or irrelevant documents.

Large Language ModelsRAGRetrieval Augmentation

0 likes · 12 min read

Can Adversarial Training Make Retrieval‑Augmented Generators More Robust?

Alimama Tech

Mar 14, 2023 · Artificial Intelligence

Neural Approximate Nearest Neighbor (NANN): Open‑Source Large‑Scale Retrieval with Arbitrary Complex Models

Alibaba’s open‑source Neural Approximate Nearest Neighbor (NANN) library decouples index learning from model training, enabling any TensorFlow‑based deep model to perform high‑throughput, high‑accuracy HNSW‑based retrieval with GPU multi‑streaming, XLA acceleration, graph optimizations, and adversarial training that mitigates L2‑distance mismatch, all supported by ready‑to‑use benchmarks and demos.

Neural ANNTensorFlowadversarial training

0 likes · 7 min read

Neural Approximate Nearest Neighbor (NANN): Open‑Source Large‑Scale Retrieval with Arbitrary Complex Models

DataFunTalk

Nov 17, 2022 · Artificial Intelligence

Enhance the Visual Representation via Discrete Adversarial Training

The Alibaba AAIG team proposes Discrete Adversarial Training (DAT), which leverages VQGAN‑based discretization to generate natural‑looking adversarial samples that improve visual representation robustness and transferability across classification, self‑supervised learning, and object detection tasks without sacrificing accuracy, achieving new state‑of‑the‑art results on multiple benchmarks.

RobustnessVisual Representationadversarial training

0 likes · 12 min read

Enhance the Visual Representation via Discrete Adversarial Training

AntTech

Oct 31, 2022 · Artificial Intelligence

Automated Attacker A² for Enhancing Model Robustness in Adversarial Training

The paper presents A², an automated, parameterized attacker that dynamically adjusts perturbation methods and step sizes during adversarial training, demonstrating improved robustness across multiple benchmarks with modest computational overhead, and outlines future directions for further efficiency and effectiveness in secure AI systems.

Machine Learning SecurityNeurIPSadversarial training

0 likes · 9 min read

Automated Attacker A² for Enhancing Model Robustness in Adversarial Training

Rare Earth Juejin Tech Community

Oct 11, 2022 · Artificial Intelligence

GANomaly: Theory and Source Code Analysis

This article explains the GANomaly model for semi‑supervised anomaly detection, detailing its generator‑encoder‑discriminator architecture, loss functions, testing phase scoring, and provides annotated PyTorch source code to help readers implement and understand the approach.

Encoder-DecoderGaNPyTorch

0 likes · 15 min read

GANomaly: Theory and Source Code Analysis

Baobao Algorithm Notes

Dec 14, 2021 · Artificial Intelligence

Five Quick Tricks to Supercharge Your Neural Network Training

This article presents five concise, widely applicable techniques—adversarial training with FGM, exponential moving average (EMA), test‑time augmentation (TTA), pseudo‑label learning, and special‑sample handling via nearest‑neighbor retrieval—to reliably improve model performance with minimal code changes.

AIEMATTA

0 likes · 7 min read

Five Quick Tricks to Supercharge Your Neural Network Training

DataFunTalk

Aug 10, 2021 · Artificial Intelligence

Practical Deep Learning Tricks: Cyclic LR, Flooding, Warmup, RAdam, Adversarial Training, Focal Loss, Dropout, Normalization, ReLU, Group Normalization, Label Smoothing, Wasserstein GAN, Skip Connections, Weight Initialization

This article presents a concise collection of practical deep‑learning techniques—including cyclic learning‑rate, flooding, warmup, RAdam, adversarial training, focal loss, dropout, various normalization methods, ReLU, group normalization, label smoothing, Wasserstein GAN, skip connections, and weight initialization—along with code snippets and references for implementation.

GaNadversarial trainingdeep learning

0 likes · 8 min read

Practical Deep Learning Tricks: Cyclic LR, Flooding, Warmup, RAdam, Adversarial Training, Focal Loss, Dropout, Normalization, ReLU, Group Normalization, Label Smoothing, Wasserstein GAN, Skip Connections, Weight Initialization

58 Tech

Jun 16, 2021 · Artificial Intelligence

Improving Text Matching Accuracy in Voice Assistants: Experiments with Siamese Networks, BERT Models, and Advanced Tricks

This article evaluates classic Siamese networks, various BERT‑based pretrained models, and several training tricks such as adversarial training, k‑fold cross‑validation, and model ensembling on both a public similarity‑sentence competition dataset and an internal voice‑assistant standard question matching dataset, ultimately raising accuracy from 97.23 % to 99.5 %.

BERTSiamese NetworkVoice Assistant

0 likes · 15 min read

Improving Text Matching Accuracy in Voice Assistants: Experiments with Siamese Networks, BERT Models, and Advanced Tricks

DataFunSummit

Mar 30, 2021 · Artificial Intelligence

Chinese Short‑Text Entity Linking: Model Design, Multitask Learning, and Experimental Results on the Qianyan Dataset

This article presents a comprehensive approach to Chinese short‑text entity linking, describing the Qianyan dataset, pipeline and end‑to‑end task formulations, sample construction, a multitask model that jointly performs entity ranking and NIL classification, various optimization techniques including confidence learning and adversarial training, and detailed experimental analysis showing state‑of‑the‑art performance.

Chinese NLPadversarial trainingconfidence learning

0 likes · 13 min read

Chinese Short‑Text Entity Linking: Model Design, Multitask Learning, and Experimental Results on the Qianyan Dataset

DataFunTalk

Feb 24, 2020 · Artificial Intelligence

Adversarial Training for Transformer‑Based Natural Language Models: Methods, Variants, and Experimental Results

This presentation reviews adversarial training techniques for transformer‑based NLP models, covering the motivation, image‑based and text‑based attack generation, standard PGD, its variants FreeAT and YOPO, the proposed FreeLB method, extensive GLUE experiments, and conclusions about robustness and future directions.

FreeLBNLPRobustness

0 likes · 18 min read

Adversarial Training for Transformer‑Based Natural Language Models: Methods, Variants, and Experimental Results

21CTO

Oct 9, 2017 · Artificial Intelligence

How Wukong’s AI Porn Detection System Achieves 99.5% Accuracy

This article explains the challenges of image‑based porn detection, details the multi‑label classification approach of the Wukong system, and reveals the deep‑learning techniques—including CNN evolution, transfer learning, loss functions, adversarial training, and GAN‑based data augmentation—that enable over 99.5% accuracy with massive daily request volumes.

CNNGaNTransfer Learning

0 likes · 18 min read

How Wukong’s AI Porn Detection System Achieves 99.5% Accuracy