ICLR 2027
Score statistics shows the whole conference and one primary area. Find papers & rankings looks up a submission by number or title.
Enter an OpenReview number or a paper title. Choose a row to see its conference rank beside its rank in its own primary area. Primary area and status shorten the list only; they do not change rank denominators. Author names appear after decisions.
Results 42,818 · Page 1 of 857
No. 37553learning on time series and dynamical systemsOpenReview
Shallow autoencoders can detect multivariate time series anomalies beyond global PCA by learning normal conditional relations that reshape low-residual regions and separate relation-violating inputs.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 34 / 1,102 · Tie · top 3.1%
No. 27873foundation or frontier models, including LLMsOpenReview
TOMC builds memory for long interaction histories on CPU by executing state, relation, and count operations and packing the results with selected source text; it outperforms six baselines on BEAM with three strong LLM...
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 1,101 / 6,357 · Tie · top 17%
No. 50145other topics in machine learning (i.e., none of the above)OpenReview
BLADE learns N:M sparsity masks block-by-block with error compensation and weight re-optimization, matching end-to-end learned mask quality (MaskLLM) on a single GPU at ~56x less compute.
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 144 / 1,080 · Tie · top 13%
No. 30069applications to physical sciences (physics, chemistry, biology, etc.)OpenReview
WindWeaver uses coarse winds to guide cross-time feature exchange in flow matching, improving multivariate weather super-resolution.
1.33
Average
Whole conference 36,537 / 40,641 · Tie · top 90%
Primary area 1,691 / 1,919 · Tie · top 88%
No. 51840optimizationOpenReview
This paper introduces a pretrain-then-calibrate framework for safe contextual optimization.
2.50
Average
Whole conference 6,157 / 40,641 · Tie · top 15%
Primary area 277 / 1,542 · Tie · top 18%
No. 18027applications to computer vision, audio, language, and other modalitiesOpenReview
We further combine diffusion with DETR through self-trajectory denoising from a noisy encoder prior, which reduces exposure bias and, with DETR training practices, surpasses DINO and DiffuDINO on COCO and LVIS under t...
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,280 / 5,224 · Tie · top 44%
No. 45403probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)OpenReview
We identify two efficiency failures that survive recalibration and introduce pooled cost projection, which preserves coverage, bounds added labels, and improves set size under matched budgets in the evaluated tasks.
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 141 / 853 · Tie · top 17%
No. 36384applications to computer vision, audio, language, and other modalitiesOpenReview
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,280 / 5,224 · Tie · top 44%
No. 57962applications to computer vision, audio, language, and other modalitiesOpenReview
A novel framework called GASplat, which reconstructs static scene with sparse input containing distractors.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,280 / 5,224 · Tie · top 44%
No. 7306neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)Desk rejectOpenReview
We treat a model's parameters as computable physical particles and other physical quantities, model their changes across iterations with mathematical and physical laws, and thereby derive the parameter update rule wit...
No score yet
No. 21979learning on graphs and other geometries & topologiesOpenReview
We show that GNNs are not invariant to node IDs althought they should, and propose a method to achieve invariance to node IDs.
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 618 / 912 · Tie · top 68%
No. 43515foundation or frontier models, including LLMsOpenReview
pipeline for comprehensive image captioning
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 4,597 / 6,357 · Tie · top 72%
No. 53378reinforcement learningWithdrawnOpenReview
No score yet
No. 59345generative modelsDesk rejectOpenReview
Interactive World Model through training free methods
No score yet
No. 53111applications to robotics, autonomy, planningOpenReview
We introduce SimpleARM, a training-free, instruction-conditioned memory that maintains and selectively retrieves task-relevant state, including entity relations, progress, and trajectories.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 109 / 2,157 · Tie · top 5.1%
No. 11316applications to computer vision, audio, language, and other modalitiesOpenReview
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 948 / 5,224 · Tie · top 18%
No. 45368foundation or frontier models, including LLMsOpenReview
Across four agent memory frameworks, we trace selective deletion from deletion operations through stored memory and context to task actions, examining the relationships between these stages.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,537 / 6,357 · Tie · top 40%
No. 6350applications to computer vision, audio, language, and other modalitiesOpenReview
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 948 / 5,224 · Tie · top 18%
No. 37926learning theoryOpenReview
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 602 / 1,308 · Tie · top 46%
No. 312reinforcement learningOpenReview
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 882 / 2,268 · Tie · top 39%
No. 37792other topics in machine learning (i.e., none of the above)OpenReview
1.33
Average
Whole conference 36,537 / 40,641 · Tie · top 90%
Primary area 957 / 1,080 · Tie · top 89%
No. 55293foundation or frontier models, including LLMsOpenReview
3.00
Average
Whole conference 291 / 40,641 · Tie · top 0.72%
Primary area 51 / 6,357 · Tie · top 0.8%
No. 23866learning on time series and dynamical systemsOpenReview
We propose Pair-Aware Contextual Tokens (PACT) framework, a relation representation that preserves global scene context while being aware of the target pair in Video Visual Relation Detection
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 129 / 1,102 · Tie · top 12%
No. 33186interpretability and explainable AIOpenReview
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 982 / 1,701 · Tie · top 58%
No. 8349learning theoryOpenReview
We show that strategy collapse is a fundamental failure mode of RLVR, explain its mechanism through optimization dynamics and information theory, and propose Mesh Learning to prevent it.
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 925 / 1,308 · Tie · top 71%
No. 23995unsupervised, self-supervised, semi-supervised, and supervised representation learningOpenReview
A lightweight class-incremental learning method that statistically models class-conditional patch distributions in the representation space.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 769 / 2,044 · Tie · top 38%
No. 31441applications to computer vision, audio, language, and other modalitiesOpenReview
KFTP makes video depth predictions more stable over time by carrying information from previous frames into the current frame with uncertainty-aware Kalman fusion.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 228 / 5,224 · Tie · top 4.4%
No. 45512foundation or frontier models, including LLMsOpenReview
We introduce the first linguistically and behaviorally motivated fine-tuning framework for identifying beneficial code-switched reasoning behaviors in large language models and teaching these models to code-switch mor...
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,537 / 6,357 · Tie · top 40%
No. 34453neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)OpenReview
Global coordination improves agents' capability for automated theorem proving
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 327 / 531 · Tie · top 62%
No. 47625applications to physical sciences (physics, chemistry, biology, etc.)OpenReview
ScaleComposer combines relational representations with scale-resolved residual learning to predict scalar air-speed fields for configurations with more sources than seen during training.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 81 / 1,919 · Tie · top 4.2%
No. 8265applications to physical sciences (physics, chemistry, biology, etc.)OpenReview
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 1,267 / 1,919 · Tie · top 66%
No. 27125foundation or frontier models, including LLMsOpenReview
We introduce RATTS, a reliability-aware test-time scaling method that allocates solution-generation compute only to useful, complementary reasoning concepts.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 2,537 / 6,357 · Tie · top 40%
No. 58919other topics in machine learning (i.e., none of the above)OpenReview
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 381 / 1,080 · Tie · top 35%
No. 41005alignment, fairness, safety, privacy, and societal considerationsOpenReview
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 427 / 3,113 · Tie · top 14%
No. 10797alignment, fairness, safety, privacy, and societal considerationsOpenReview
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 2,185 / 3,113 · Tie · top 70%
No. 48334datasets and benchmarksOpenReview
FactorBench evaluates whether advances in automated factor mining yield signals that generalize, contribute distinct information, and improve portfolio performance after trading costs.
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 509 / 2,977 · Tie · top 17%
No. 33362applications to computer vision, audio, language, and other modalitiesOpenReview
A frozen SAM 3, guided by image-learned forensic evidence and a causal memory bank, localizes video forgeries in space and time, scored by a new gated protocol.
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 948 / 5,224 · Tie · top 18%
No. 12870learning on graphs and other geometries & topologiesOpenReview
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 618 / 912 · Tie · top 68%
No. 35017unsupervised, self-supervised, semi-supervised, and supervised representation learningOpenReview
We introduce a path law regularizer for Joint-Embedding World Models (JEPA) that eliminates static latent collapse and significantly improves long-horizon visual planning.
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 1,389 / 2,044 · Tie · top 68%
No. 40924foundation or frontier models, including LLMsOpenReview
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 4,597 / 6,357 · Tie · top 72%
No. 16680unsupervised, self-supervised, semi-supervised, and supervised representation learningOpenReview
CFF+MAE introduces block-wise ViT pretraining that jointly learns contrastive and masked reconstruction objectives without end-to-end backpropagation, enabling strong representations with much lower training memory.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 769 / 2,044 · Tie · top 38%
No. 52598learning on graphs and other geometries & topologiesOpenReview
We developed two boundary detection methods for high-dimensional data, and we applied them to spatial transcriptomics data and LLM activations.
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 331 / 912 · Tie · top 36%
No. 44775unsupervised, self-supervised, semi-supervised, and supervised representation learningOpenReview
Zero-shot cross-city mobility generation by transferring activity patterns as discrete, city-independent activity prototypes, using only the target city's static map and residential distribution.
1.33
Average
Whole conference 36,537 / 40,641 · Tie · top 90%
Primary area 1,821 / 2,044 · Tie · top 89%
No. 21394probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)OpenReview
We propose a novel framework for evaluating agent from relative feedback that represents the collective preferences of the entire population.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 40 / 853 · Tie · top 4.7%
No. 48323reinforcement learningOpenReview
The value signal determines where along the flow guidance works, while the number of guidance determines how well the guidance budget is used.
2.67
Average
Whole conference 1,905 / 40,641 · Tie · top 4.7%
Primary area 98 / 2,268 · Tie · top 4.3%
No. 51037interpretability and explainable AIOpenReview
Time series attribution has no convention-free answer; what does is how much of a declared value function an explanation of a given size carries. We measure that distortion-rate curve, refute locality for low rank, an...
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 505 / 1,701 · Tie · top 30%
No. 27514alignment, fairness, safety, privacy, and societal considerationsOpenReview
2.00
Average
Whole conference 16,177 / 40,641 · Tie · top 40%
Primary area 1,157 / 3,113 · Tie · top 37%
No. 40996datasets and benchmarksOpenReview
AWDM evaluates whether computer-use agents can keep making progress as user intent evolves without acting beyond what the user has decided.
2.33
Average
Whole conference 6,809 / 40,641 · Tie · top 17%
Primary area 509 / 2,977 · Tie · top 17%
No. 4754applications to computer vision, audio, language, and other modalitiesOpenReview
We propose SAG, a structured retrieval architecture that activates latent hyperedges over event–entity relations with query-time SQL joins.
2.25
Average
Whole conference 15,377 / 40,641 · Tie · top 38%
Primary area 2,178 / 5,224 · Tie · top 42%
No. 39298applications to physical sciences (physics, chemistry, biology, etc.)OpenReview
We introduce CellKAFC, a knowledge-aware flow-correction framework that integrates gene-level biological priors with flow correction.
1.67
Average
Whole conference 28,855 / 40,641 · Tie · top 71%
Primary area 1,267 / 1,919 · Tie · top 66%