SOTA – AI Community
SOTA-AI Editorials
Write-ups of olympiad tasks by members of the SOTA community, each with hints you can open one at a time before the full solution.
The olympiad's official statement and scoring win wherever an editorial differs. Results are as their authors measured them.
| Task | Round | Result | By |
|---|---|---|---|
| Day 2 Audio (Sped-up Speech Detection) | 2026 · Day 2 | 0.997 AUC-ROC on the private leaderboard | Luvidi Alghari |
| Creatures of the Static | 2026 · Day 2 | 0.978 Micro-F1 on the private dataset | Luvidi Alghari |
| Task | Round | Result | By |
|---|---|---|---|
| The Clothing Sale | 2024 · BdAIO 2024 problems (round not stated) | — | Rishikesh Muralimohan |
| Task | Round | Result | By |
|---|---|---|---|
| Applied Problem Solving: Loan Default Prediction | 2025 · National Qualifier | No score measured because the official CSV files were inaccessible; the script provides a runnable candidate pipeline. · Not measured; see result for the limitation. | Gonçalo Franco |
| Task | Round | Result | By |
|---|---|---|---|
| User Intent Recognition in Zhihu Scenarios | 2026 · Round 2 (China Stage) | Candidate weighted F1 not measured because task data and supplied BERT checkpoint were unavailable; organizer reference B is 0.8154, not this result. · Not measured; see result for the limitation. | Gonçalo Franco |
| Embodied Intelligence Sim2Real State Prediction Based on NVIDIA Isaac Sim | 2026 · Round 2 (China Stage) | Candidate score not measured because task data were unavailable; organizer reference B is 0.7861, not this result. · Not measured; see result for the limitation. | Gonçalo Franco |
| Product · Sum | 2026 · Round 2 (China Stage) | Candidate accuracy not measured because task data were unavailable; organizer reference B is 0.9600, not this result. · Not measured; see result for the limitation. | Gonçalo Franco |
| Maze Information Prediction | 2026 · Round 2 (China Stage) | Candidate score not measured because task data were unavailable; organizer reference B is 0.8653, not this result. · Not measured; see result for the limitation. | Gonçalo Franco |
| Compound Word Segmentation | 2025 · Round 2 (China Stage, national final) | 0.862135 exact-span F1 on a related IOAI GAITE holdout; NOAI score and data identity remain unverified. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| Chemical Reaction Kinetics Simulation | 2025 · Round 2 (China Stage, national final) | No candidate score measured because organizer data and grader were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Synthetic Speech Detector | 2025 · Round 2 (China Stage, national final) | No NOAI validation or leaderboard score measured; related IOAI dataset identity is unverified. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| Grid Collage Classification | 2025 · Round 2 (China Stage, national final) | No candidate accuracy measured because labeled collage targets were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Predicting the Shooting Percentage of Basketball Stars | 2024 · Round 2 (China Stage) | 0.5978 default-threshold OOF accuracy on a community mirror; threshold-tuned OOF is optimistic, and official hidden-test score is unmeasured. · Not measured; see result for the limitation. | Gonçalo Franco |
| Real or Fake Image Recognition Task | 2024 · Round 2 (China Stage) | Local holdout accuracy 0.7205 and implied task formula score 0.7279; participant-reported leaderboard accuracy is not this result. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| News Text Classification Task | 2024 · Round 2 (China Stage) | 0.9724 duplicate-aware OOF macro-F1 on a community mirror; official hidden-test score is unmeasured. · Not measured; see result for the limitation. | Gonçalo Franco |
| Solving the Pendulum Motion with Missing Data | 2024 · Round 2 (China Stage) | 0.000270 rad RMSE on the visible mirrored waveform (fit diagnostic only); official parameter score is unmeasured. · Not measured; see result for the limitation. | Gonçalo Franco |
| Task | Round | Result | By |
|---|---|---|---|
| The Five Artists | 2025 · Final Round | No accuracy measured because competition image files were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Heart Disease Risk Prediction | 2025 · Problem 1 (round not stated) | No F1 measured because competition CSV files were unavailable; no Kaggle submission result is claimed. · Not measured; see result for the limitation. | Gonçalo Franco |
| Task | Round | Result | By |
|---|---|---|---|
| Classify Gases from Multimodal Information | 2025 · Competition | No score measured by this solution; a participant write-up reported 0.984 in the open track, which is not this result. · Not measured; see result for the limitation. | Gonçalo Franco |
| Task | Round | Result | By |
|---|---|---|---|
| LeetCode Difficulty Prediction from Solution Descriptions | 2025 · IOAI Team Selection Test | 0.37509 F1-Score on the released test set · Unverified | Aboubacar Diarra |
| Task | Round | Result | By |
|---|---|---|---|
| Decision Localisation | 2026 · Final (Stage III) | Not measured; the image archive and model are not present in this workspace · Unverified | Gonçalo Franco |
| The Void | 2026 · Final (Stage III) | 5/5 trigger pairs recovered · 100/100 | Gonçalo Franco |
| Pixels | 2026 · Final (Stage III) | 63.45% accuracy · 100/100 | Gonçalo Franco |
| Hidden Categories | 2026 · Final (Stage III) | 0.483 mean IoU · 100/100 | Gonçalo Franco |
| Painter's Optimiser | 2026 · Stage II | 0.004927 mean MSE · 100/100 | Gonçalo Franco |
| Token Predictor | 2026 · Stage II | Mean balanced accuracy 0.684145 and 100/100 on 99 released validation examples; hidden-test performance is unknown. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Decision Trees | 2026 · Stage II | 66.0/100 on released collection A (55 of 100 datasets matched or beat baseline); full credit was not reached. · 66.0/100 (see result for validation and provenance) | Gonçalo Franco |
| Colourisation with a GAN | 2026 · Stage II | Generator-only candidate score unverified because the checkpoint and LPIPS runtime were unavailable; a validation-target-fitted diagnostic is excluded. · Not measured; see result for the limitation. | Gonçalo Franco |
| Inpainting with Implicit Neural Representations | 2025 · Final (Stage III) | 100/100 after official rounding on released validation (31.84 dB PSNR and 97.871% mask accuracy); hidden-test score unknown. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Machine (Un)learning | 2025 · Final (Stage III) | The participant notebook recorded 99.0049/100 on released target class 9; not freshly reproduced, and hidden-target generalization is unknown. · 99.0049/100 (see result for validation and provenance) | Gonçalo Franco |
| Translation Styling | 2025 · Final (Stage III) | No BLEU or points measured because the Marian checkpoint was unavailable; candidate fine-tuning code is included. · Not measured; see result for the limitation. | Gonçalo Franco |
| Data Prototypes | 2025 · Final (Stage III) | Candidate score unmeasured because embeddings were unavailable; the notebook random baseline scored 0/100 (18.34% accuracy). · Candidate not scored; the result describes the notebook baseline. | Gonçalo Franco |
| Non-Normal Distribution | 2025 · Stage II | 100/100 locally measured on released validation using training data only; hidden-test performance is unknown. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Credit Taking – Explaining Model Decisions | 2025 · Stage II | 100/100 locally reproduced on released validation; mean L2 change 0.2138, validity 1.000, and plausibility 1.000. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Source Extraction | 2025 · Stage II | Fallback achieved nDCG@10 0.50365 and 100/100 on released validation; SGPT and hidden-test performance remain unverified. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Coin Counting Machine | 2025 · Stage I | No released-validation score measured because Google Drive data were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Hallucination Detection | 2025 · Stage I | Organizer notebook reports AUC 0.8226 and 100/100; this is not a fresh local measurement. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| ECG Signal Disorder Detection | 2025 · Stage I | 98.4413% balanced accuracy and 100/100 locally measured on released validation. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Noise in Data Labels | 2025 · Stage I | Organizer notebook reports mean balanced accuracy 0.88123 and full credit; not independently rerun here. · Full credit; see result for metric and provenance. | Gonçalo Franco |
| Hidden Subsequences | 2025 · Stage I | Organizer notebook reports validation MSE 1.0468 and 98/100; local reproduction was blocked by unavailable data. · 98/100 (see result for validation and provenance) | Gonçalo Franco |
| Ciphers | 2024 · Stage II – Final | No score or runtime measured: the cipher corpus and ground truth were unavailable; the decoder is a candidate implementation. · Not measured; see result for the limitation. | Gonçalo Franco |
| Anomaly Detection | 2024 · Stage II – Final | 91.30% accuracy, 0.97534 ROC AUC, and 100/100 points on released validation; the threshold was fit using normal training examples only. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Self-Supervised Learning (A Classifier in Two Minutes) | 2024 · Stage II – Final | The participant notebook recorded 89.46295% validation accuracy and 0.97315/1 points; not freshly reproduced because required tensors were missing. · 0.97315/1 points | Gonçalo Franco |
| Machine Translation | 2024 · Stage II – Implementation Contest | BLEU and rubric points are unmeasured because Multi30k was unavailable; the model and training/evaluation utilities are implemented. · Not measured; see result for the limitation. | Gonçalo Franco |
| Adversarial Attacks | 2024 · Stage I | Criterion 48.65 on released validation, above the full-credit threshold; SSIM was reconstructed rather than computed with scikit-image. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| Imbalanced Classification | 2024 · Stage I | 100/100 in a local compact-CNN experiment; a clean standalone rerun and runtime check remain unverified. · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Dependency Parsing | 2024 · Stage I | No parser score measured: the required HerBERT checkpoint was unavailable; the report covers data analysis and candidate approaches. · Not measured; see result for the limitation. | Gonçalo Franco |
| Colour Quantisation | 2024 · Stage I | Mean validation cost 6,950.8 with 37 colors, locally measured below the 8,000 full-credit threshold (100/100). · 100/100 (see result for validation and provenance) | Gonçalo Franco |
| Object Tracking | 2024 · Stage I | Examples only: 2/2, 2/2, and 1/2 across the three subtasks; released validation files were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Pruning | 2024 · Stage I | 0.9721 on released validation and full task credit (1.5/1.5 points), checked with the official validation script. · 1.5/1.5 points | Gonçalo Franco |
| Riddles | 2024 · Stage I | No score measured: the required riddle corpus, morphology data, and embeddings were unavailable. · Not measured; see result for the limitation. | Gonçalo Franco |
| Task | Round | Result | By |
|---|---|---|---|
| Section 1: Multiple-Choice Questions | 2026 · Final Assessment | No score measured because there is no official key or public grader; several released questions are ambiguous. · Not measured; see result for the limitation. | Gonçalo Franco |
| Gradient Boosting from Scratch | 2026 · Final Assessment | Synthetic CPU checks passed; no task-level score or organizer grading result is available. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| The Segmentation Fracture (Debugging U-Net) | 2026 · Final Assessment | Forward-pass and metric checks passed on synthetic CPU inputs; no dataset-based or organizer score was measured. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| The Attention Mechanism (Building the "Brain") | 2026 · Final Assessment | Synthetic CPU checks passed; no task-level score, training run, or organizer grading result is available. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| Preliminary Round Assessment (300 MCQs) | 2026 · Preliminary Round | No score measured because no official answer key or grader was released; the derived key flags uncertain items. · Not measured; see result for the limitation. | Gonçalo Franco |
| Section 1: Multiple-Choice Questions | 2025 · Final | 20/20 matches the published answer key; this is not an organizer-submitted score. · 20/20 published answers matched; not an organizer-issued score. | Gonçalo Franco |
| Linear Regression Using Gradient Descent | 2025 · Final | The released example output exactly matches the official notebook; no organizer-issued mark or hidden grader result is available. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| K-Means Clustering | 2025 · Final | The released example centroids exactly match the official notebook; no organizer-issued mark or hidden grader result is available. · No points score reported; see result for the measured metric. | Gonçalo Franco |
| Retrieval Augmented Generation | 2025 · Final | The expected source order and grounded response matched the published reference; no public rubric or organizer score is available. · No points score reported; see result for the measured metric. | Gonçalo Franco |