{"total":12,"items":[{"citing_arxiv_id":"2606.29975","ref_index":6,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Atompack: A Storage and Distribution Layer for Read-Heavy Atomistic ML Training Datasets","primary_cat":"cs.LG","submitted_at":"2026-06-29T08:50:15+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Atompack delivers 96x faster shuffled reads and 79% smaller artifacts than ASE LMDB baselines for complete-record atomistic ML training workloads.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.21632","ref_index":35,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Fine-Tuning a Universal Machine-Learned Interatomic Potential for Oxygen Plasma Interactions with WS$_2$","primary_cat":"cond-mat.mtrl-sci","submitted_at":"2026-06-19T17:36:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Pretrained UMA model reproduces chemisorbed S and O coverage under 15 eV O+ and O2+ bombardment on WS2 without fine-tuning; fine-tuning lowers energy MAE to 4.5e-3 eV/atom and force MAE to 0.076 eV/Å.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.00794","ref_index":11,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Benchmark Dataset for Catalysis on 2D MXenes","primary_cat":"cond-mat.mtrl-sci","submitted_at":"2026-05-30T16:18:17+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"A benchmark dataset of 60,000 DFT calculations on 2D MXenes is created and used to train MLIPs achieving ~1000-4000x CPU speedup with ~10 meV/A force and ~1 meV/atom energy accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25717","ref_index":12,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"FLOATBench: A Dataset and Benchmark for Floating Offshore Wind Turbine Tower Fatigue","primary_cat":"cs.AI","submitted_at":"2026-05-25T11:18:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":8.0,"formal_verification":"none","one_line_summary":"FLOATBench is a tabular benchmark dataset with 582,120 fatigue labels from 19,404 OpenFAST simulations of three 22 MW FOWT towers, featuring alpha-shape regime partitioning and three evaluation protocols for surrogate models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.14154","ref_index":23,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"TSAgent: An Agentic Workflow for Autonomous Transition State Search","primary_cat":"physics.chem-ph","submitted_at":"2026-05-13T22:08:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"TSAgent automates transition state searches at DFT accuracy via an agentic loop, reaching 83% success on 100 OC20NEB examples and 70% on 10 held-out cases versus 73% for human experts.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.13788","ref_index":117,"ref_count":2,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs","primary_cat":"cs.LG","submitted_at":"2026-05-13T17:08:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Force-aware Neural Tangent Kernels combined with chunked acquisition provide scalable and distribution-robust active learning for MLIPs, outperforming baselines on OC20 and remaining competitive on other benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.09394","ref_index":14,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Systematic Fine-Tuning of MACE Interatomic Potentials for Catalysis","primary_cat":"physics.chem-ph","submitted_at":"2026-05-10T07:43:42+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Fine-tuned MACE MLIPs achieve lower mean absolute errors on catalytic reaction energies and barriers than from-scratch models, with a large fine-tuned model performing best on both metallic and oxide systems including out-of-distribution cases.","context_count":1,"top_context_role":"dataset","top_context_polarity":"use_dataset","context_text":"improved flexibility and reactivity over classical molecular dynamics (MD) force fields. Recent equivariant message passing MLIP architectures can efficiently learn atom environment representations and have greatly improved accuracy and transferability over previous approaches [11-13]. For heterogeneous catalysis, the reliance on large datasets to train MLIPs such as the Open Catalyst datasets [14] \"out of the box\" might be a limitation, as performance remains heavily dependent on data availability and quality, which may restrict their generalizability when determining bothE r and Ea Wood, et al. [15] have used a mixture of experts model to train a family of universal models (UMA) on a combined dataset comprised of OC22 (oxide electrocatalysts) [16],"},{"citing_arxiv_id":"2605.07714","ref_index":9,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Selectivity- and Activity-Aware Catalyst Descriptors for CO$_2$ Hydrogenation on Alloy Nanocatalysts using Machine-Learned Force Fields","primary_cat":"cond-mat.mtrl-sci","submitted_at":"2026-05-08T13:17:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A facet-resolved adsorption energy distribution method with ML force fields identifies active and methanol-selective alloy nanocatalyst surfaces for CO2 hydrogenation.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"; Yu, J.; Fang, C.; Jian Sun Disentangling the activity-selectivity trade-off in CO2 hydrogenation to methanol.Chem2026, 102942. (8) Benavides-Hern' andez, J.; Dumeignil, F. From Characterization to Discovery: Artificial Intelligence, Machine Learning and High-Throughput Experiments for Heterogeneous Catalyst Design.ACS Catalysis2024,14, 11749-11779. (9) Ortega, C.; Otyuskaya, D.; Ras, E.; Virla, L. D.; Patience, G. S.; Dathe, H. Experi- mental methods in chemical engineering: High throughput catalyst testing - HTCT. The Canadian Journal of Chemical Engineering2021,99, 1288-1306. (10) Dongapure, P.; Tekawadia, J.; Thundiyil, S.; Caha, I.; Deepak, F. L.; Devi, R. N. Mech- anistic Insights into Near Ambient Pressure Activity of Intermetallic NiZn/TiO 2 Cat-"},{"citing_arxiv_id":"2605.03964","ref_index":112,"ref_count":2,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Pretrained Model Representations as Acquisition Signals for Active Learning of MLIPs","primary_cat":"cs.LG","submitted_at":"2026-05-05T16:48:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Kernels from pretrained MLIP latent spaces outperform standard acquisition methods in active learning for reactive chemistry, reducing required labels by 38% for energy error and 28% for force error.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.24563","ref_index":62,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"Enhancing molecular dynamics with equivariant machine-learned densities","primary_cat":"physics.chem-ph","submitted_at":"2026-04-27T14:49:53+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DenSNet learns the Hohenberg-Kohn map to electron density with equivariant networks and delta-learning, then maps density to energy, producing stable MD trajectories whose infrared spectra match experiment and DFT on ethanol, ethanethiol, resorcinol, and polythiophene oligomers.","context_count":1,"top_context_role":"background","top_context_polarity":"unclear","context_text":"[61] Lowik Chanussot, Abhishek Das, Siddharth Goyal, Thibaut Lavril, Muhammed Shuaibi, Morgane Riviere, Kevin Tran, Javier Heras-Domingo, Caleb Ho, Weihua Hu, Aini Pal- izhati, Anuroop Sriram, Brandon Wood, Junwoong Yoon, Devi Parikh, C. Lawrence Zitnick, and Zachary Ulissi. Open catalyst 2020 (OC20) dataset and community chal- lenges.ACS Catalysis, 11:6059-6072, 2021. doi: 10.1021/acscatal.0c04525. [62] Daniel S Levine, Muhammed Shuaibi, Evan Walter Clark Spotte-Smith, Michael G Tay- lor, Muhammad R Hasyim, Kyle Michel, Ilyes Batatia, G' abor Cs' anyi, Misko Dzamba, Peter K Eastman, et al. The Open Molecules 2025 (OMol25) dataset, evaluations, and models, 2025. [63] Stefan Ganscha, Oliver T Unke, Daniel Ahlin, Hartmut Maennel, S Kashubin, and Klaus-Robert M¨ uller."},{"citing_arxiv_id":"2604.21830","ref_index":4,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"GFlowState: Visualizing the Training of Generative Flow Networks Beyond the Reward","primary_cat":"cs.LG","submitted_at":"2026-04-23T16:22:55+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GFlowState introduces interactive visualizations such as trajectory node-link diagrams and transition heatmaps to make GFlowNet training dynamics observable for debugging and quality assessment.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"lenges such as climate change and global health by advancing battery technologies, electrocatalysts for hydrogen generation [23], and the design of therapeutic molecules such as antibodies for combating in- fectious diseases [5]. Large-scale initiatives in machine learning for scientific discovery have demonstrated the potential of data-driven ap- proaches to explore vast chemical and material spaces [4]. However, these spaces are combinatorially large, making an exhaustive search infeasible, and motivating algorithms that can efficiently explore and prioritize promising candidates. Generative Flow Networks(GFlowNets or GFNs for short) [1] were designed to tackle this task by learning to sample objects proportionally to a reward function. This principle, combined with the generaliza-"},{"citing_arxiv_id":"2604.09130","ref_index":2,"ref_count":1,"confidence":0.88,"is_internal_anchor":false,"paper_title":"EquiformerV3: Scaling Efficient, Expressive, and General SE(3)-Equivariant Graph Attention Transformers","primary_cat":"cs.LG","submitted_at":"2026-04-10T09:12:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"EquiformerV3 advances prior equivariant transformers via efficiency tweaks and new activations, achieving SOTA on OC20, OMat24, and Matbench Discovery with DeNS auxiliary training.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}