A Semi-supervised Molecular Learning Framework for Activity Cliff Estimation 1. This paper introduces SemiMol, a novel semi-supervised learning (SSL) framework designed to enhance molecular property predictions in the presence of activity cliffs, a challenging scenario where structurally similar molecules exhibit vastly different properties. The method leverages unannotated data to improve model performance in low-data situations. 2. SemiMol employs an instructor model to evaluate the accuracy and trustworthiness of pseudo-labels generated from unannotated data. This addresses a critical issue in SSL where pseudo-labels can be unreliable due to differences between labeled and unlabeled data distributions. 3. The framework incorporates a self-adaptive curriculum learning algorithm, which progressively moves the target model towards harder samples at a controllable pace. This approach prevents the accumulation of errors from unreliable pseudo-labels and ensures robust training. 4. Extensive experiments on 30 activity cliff datasets demonstrate that SemiMol significantly outperforms state-of-the-art pretraining and SSL methods, achieving an average improvement of 26.53% in RMSE. This highlights its effectiveness in capturing chemical and biological information for accurate activity cliff estimation. 5. The study also investigates the limitations of self-supervised graph pretraining in activity cliff estimation, finding that pretraining benefits are often negligible or even negative. This suggests that SSL methods like SemiMol may be more effective in such scenarios. 📜Paper: arxiv.org/abs/2601.04507v1 #MachineLearning #SemiSupervisedLearning #MolecularPropertyPrediction #ActivityCliffs #DrugDiscovery

✈️🇸🇬 to #ICLR 2025 🔥🔥🔥 at the iconic city of #Singapore participating in The Thirteenth International Conference on Learning Representations, one of the 4 main #machinelearning #ai conferences worldwide, with Dr @josesanchezhb This year with promissing Invited talks by @dawnsongtweets Song-Chun Zhu @danqi_chen @zicokolter @YiMaTweets @_rockt and 44 workshops, 3827 papers, orals, posters, socials and many more, featuring @SchmidhuberAI @SLapuschkin @lifu_huang @Yoshua_Bengio @sea_snell @wellingmax @svlevine @pabbeel to name a very limited few Thanks to the ICLR organizers: @yisongyue @cvondrick @yuqirose @animesh_garg @orussakovsk @pcastr @francescazfl @savvyRL @fredahshi @SchwinnLeo Jonas Köhler and many others, including the 10s of sponsors like: @Microsoft @AIatMeta @Google @amazon @Oracle @Huawei @Apple @UnitreeRobotics and many others for making it possible one more year. See you all! PD: We will be hosting two @_Qubic_ AGI dinners on the 24th & 25th seats are very limited but DM if you are interested #Artificialntelligence #AI #AGI #RepresentationLearning #FeatureLearning #UnsupervisedLearning #SemiSupervisedLearning #SupervisedLearning #MetricLearning #KernelLearning #SparseCoding #DimensionalityExpansion #HierarchicalModels #OptimalTransport #DeepLearningTheory #Planning #ReinforcementLearning #ComputerVision #NLP #AudioProcessing #SpeechRecognition #Robotics #Neuroscience #Biology #ClimateScience #Sustainability #Fairness #AIethics #Safety #Privacy #Interpretability #ExplainableAI #Visualization #Optimization #TheEndOfKnowledge #Artificiology

🔥 Read our Paper 📚 Application of Semi-Supervised Learning Model to Coal Sample Classification 🔗 mdpi.com/2076-3417/14/4/1606 👨‍🔬 by Dongming Wang,Li Xu,Wei Gao,Hongwei Xia,Ning Guo andXiaohan Ren #semisupervisedlearning #laserinducedbreakdownspectrum #classification #coal @ShandongU

S-MolSearch: 3D Semi-supervised Contrastive Learning for Bioactive Molecule Search • S-MolSearch introduces a groundbreaking semi-supervised framework combining 3D molecular structures and affinity data, setting new standards in ligand-based virtual screening (LBVS). • Leveraging inverse optimal transport, the method trains two encoders: one for labeled data and one for both labeled and unlabeled data. This design allows efficient knowledge transfer, dramatically improving performance on scarce labeled datasets. • S-MolSearch achieves state-of-the-art results on benchmarks like DUD-E and LIT-PCBA, with a 49% improvement in BEDROC and a 30% enhancement in enrichment factor (EF) over previous best methods. • Unlike structure-based methods, S-MolSearch does not rely on protein target structures, making it suitable for challenging scenarios such as disordered proteins and cell-based assays. • The method uses advanced data curation from ChEMBL, processing nearly 600,000 protein-molecule pairs while employing a molecular pretraining backbone (Uni-Mol) for robust representation learning. • Extensive ablation studies confirm the effectiveness of its components, including soft labeling, regularization, and pretraining, ensuring adaptability to both labeled and unlabeled data. • The framework demonstrates its power in the zero-shot setting and shows potential in few-shot scenarios, significantly outperforming traditional methods in identifying bioactive molecules. @guolin_ke 💻Code: github.com/s-molsearch-team/… 📜Paper: arxiv.org/pdf/2409.07462 #VirtualScreening #DrugDiscovery #Bioinformatics #MachineLearning #SemiSupervisedLearning #OptimalTransport #MolecularSearch

Instructor-inspired Machine Learning for Robust Molecular Property Prediction 1. InstructMol introduces an innovative framework for semi-supervised learning (SSL) in molecular property prediction, leveraging an instructor model to evaluate the reliability of pseudo-labels, enabling efficient use of large-scale unlabeled data. 2. Unlike traditional pretrain-finetune paradigms, InstructMol avoids domain transfer gaps by employing a pseudo-labeling mechanism tailored for specific tasks, ensuring high compatibility with both labeled and unlabeled datasets. 3. A key innovation of InstructMol is its integration of an instructor model as a “critic,” guiding the target model to assign appropriate attention to labeled and pseudo-labeled data, significantly reducing noise and improving prediction accuracy. 4. Extensive benchmarks on MoleculeNet and out-of-distribution (OOD) datasets reveal that InstructMol surpasses state-of-the-art SSL approaches, achieving up to a 9.98% reduction in RMSE for regression tasks and notable improvements in classification metrics. 5. In real-world applications, InstructMol accurately predicted properties of nine newly patented drug molecules, validating its robustness against experimental Ki values for the 5-HT1A receptor with minimal error. 6. InstructMol is highly effective in low-data scenarios, achieving significant prediction improvements with only 0.1% of labeled training data, showcasing its robustness in addressing data scarcity challenges. 7. The framework is compatible with various machine learning architectures, including GCNs, GATs, and GINs, offering broad applicability across diverse molecular modeling tasks. 8. Future directions include the development of bespoke self-supervised learning techniques that synergize with InstructMol, further enhancing its capability to address complex biochemical challenges. @WUFang40615703 @LupinLSY 📜Paper: openreview.net/pdf?id=j7sw0n… #MolecularPrediction #SemiSupervisedLearning #MachineLearning #DrugDiscovery #Bioinformatics

🔔🔔🔔 #MDPIfutureinternet [New Published Papers in 2024] Title: Advancing Additive Manufacturing Through #MachineLearning Techniques: A State-of-the-Art Review mdpi.com/1999-5903/16/11/419 #supervisedlearning #semisupervisedlearning #reinforcementlearning

'You can’t handle the (dirty) truth: Data-centric Insights improve Pseudo-Labeling' by Nabeel Seedat, Nicolas Huynh, Fergus Imrie, Mihaela van der Schaar Action Editor: Sergio Escalera data.mlr.press/assets/pdf/v0… #PseudoLabeling #SemiSupervisedLearning #DataCharacterization

BrainExpand🧠 - Exploratory AI Pipeline Overview DataSets: DataCollection (Python, R, Apache Kafka, Google Colab, Kaggle) ➡ DataCleaning (Pandas, OpenRefine, Trifacta) ➡ DataAugmentation (imgaug, Albumentations) ➡ DataNormalization (scikit-learn, TensorFlow) ➡ DataSplitting (scikit-learn, train_test_split) Algorithms: SupervisedLearning (scikit-learn, TensorFlow, Azure ML) ➡ UnsupervisedLearning (scikit-learn, Keras, IBM Watson) ➡ SemiSupervisedLearning (scikit-learn, PyTorch, Google Colab) ➡ ReinforcementLearning (OpenAI Gym, TensorFlow, Azure ML) NeuralNetworks: Layers (Keras, PyTorch) ➡ ActivationFunctions (TensorFlow, Keras) ➡ WeightInitialization (PyTorch, Keras) ➡ Backpropagation (TensorFlow, PyTorch) ➡ GradientDescent (scikit-learn, TensorFlow) Overfitting: Regularization (Keras, TensorFlow) ➡ Dropout (Keras, TensorFlow) ModelEvaluation: CrossValidation (scikit-learn) ➡ Metrics (scikit-learn, TensorFlow) FeatureEngineering: FeatureSelection (scikit-learn, Boruta) ➡ FeatureExtraction (Pandas, scikit-learn) ModelTraining: Training (TensorFlow, PyTorch, Databricks) ➡ Validation (scikit-learn) ➡ Testing (scikit-learn) HyperparameterTuning: GridSearch (scikit-learn) ➡ RandomSearch (scikit-learn) ➡ CrossValidation (scikit-learn) FineTuning (Keras, PyTorch) ➡ TransferLearning (TensorFlow, Keras) ModelDeployment: Docker, Kubernetes, AWS SageMaker, Google Cloud AI Platform, Azure ML ➡ Monitoring (Prometheus, Grafana, MLflow) ➡ Retraining (Kubeflow, Apache Airflow) Interpretability: SHAP, LIME, PDP Collaboration and Automation: Git, MLflow, Weights & Biases, Kubeflow, Apache Airflow #AI #DeepLearning #MachineLearning #AIUnlock #TechMystery

Researchers from @EmoryUniversity and @GeorgiaTech found that the NeST method enhances self-training with a neighborhood-regularized approach, boosting accuracy & efficiency in tasks with few labels: pubmed.ncbi.nlm.nih.gov/3833… #AI #SemiSupervisedLearning @GTResearchNews

🔔New Research by Mr. Hiromu Nakajima and Dr. Minoru Sasaki : "Text Classification Based on the Heterogeneous Graph Considering the Relationships between Documents" @ComSciMath_Mdpi #textclassification #CNN #semisupervisedlearning Access for Free: mdpi.com/2504-2289/7/4/181

Semi-Supervised Learning Roadmap with Case studies and Python code 1. Like & Repost 2. Reply 'get' 3. Follow me (so I can DM Road Map Link) #SemiSupervisedLearning #MachineLearning #DataScience #AIJourney #RoadmapToSuccess #MLCommunity #TechLearning

Semi-supervised Learning with Report-guided Pseudo-Labels for Deep Learning-based Prostate Cancer Detection using Biparametric MRI doi.org/10.1148/ryai.230031 @joeranbosma @anindox8 @radboudumc #SemisupervisedLearning #cancer #prostate

Report-guided semisupervised learning outperformed semisupervised learning in clinically significant prostate cancer detection doi.org/10.1148/ryai.230031 @joeranbosma @MaartendeRooij @radboudumc #Semisupervised #SemisupervisedLearning #AI

Semi-supervised Learning with Report-guided Pseudo-Labels for Deep Learning-based Prostate Cancer Detection using Biparametric MRI doi.org/10.1148/ryai.230031 @joeranbosma @anindox8 @MaartendeRooij #SemisupervisedLearning #AI #ML

Report-guided semisupervised learning outperformed semisupervised learning in clinically significant prostate cancer detection doi.org/10.1148/ryai.230031 @anindox8 @MaartendeRooij @radboudumc #Semisupervised #SemisupervisedLearning #ProstateCancer

Semi-supervised Learning with Report-guided Pseudo-Labels for Deep Learning-based Prostate Cancer Detection using Biparametric MRI doi.org/10.1148/ryai.230031 @anindox8 @MaartendeRooij @radboudumc #SemisupervisedLearning #ProstateCancer #ML

Three typical machine learning algorithms for flood forecasting are reviewed: #supervisedlearning, #unsupervisedlearning, and #semisupervisedlearning. #floodforecasting, #deeplearning @DaniloMcgarry @_ahmedzaidi @DamienERNST1 @Lang__Leon More: intellrobot.com/article/view…

A novel semisupervised learning method leverages clinical reports to guide voxel-level labels doi.org/10.1148/ryai.230031 @joeranbosma @anindox8 @radboudumc #SemisupervisedLearning #ProstateCancer #prostate

A novel semisupervised learning method leverages clinical reports to guide voxel-level labels doi.org/10.1148/ryai.230031 @anindox8 @MaartendeRooij @radboudumc #SemisupervisedLearning #AI #ML

Semi-supervised Learning with Report-guided Pseudo-Labels for Deep Learning-based Prostate Cancer Detection using Biparametric MRI doi.org/10.1148/ryai.230031 @joeranbosma @anindox8 @MaartendeRooij #SemisupervisedLearning #AI #ML