Task bank

Tasks

Browse the tasks by category, intro, evaluation setting.

Public set

LLM Agent Memory Architecture Design

AI Models & Agents

Design an LLM agent memory architecture that generalizes

Scientific Discovery Agent Harness Design

AI Models & Agents

Redesign a ReAct harness for long-horizon scientific discovery in unseen simulated worlds

Small-Model Math Reasoning Post-Training

AI Models & Agents

Design a distillation recipe for a 1.7B student using a fixed 8B teacher

Vision-Language Compositional Generalization Post-Training

AI Models & Agents

Design a GRPO recipe for visual counting across unseen attribute combinations and renderers

Lean 4 Proof-Search LLM agent Workflow Design

AI Models & Agents

Design a compiler-guided LLM workflow for unseen theorems under fixed inference and checking budgets

Video Point-Track Candidate Routing

AI Models & Agents

Select candidate trajectories and predict visibility under a four-switch budget on unseen videos

TriFinger Cube-Pushing Offline RL

AI Models & Agents

Train a control policy from fixed logged trajectories without simulator interaction

Multi-View Point Cloud Fusion

AI Models & Agents

Fuse noisy multi-view observations into one point cloud

10M-Scale Metadata-Filtered ANN Search

Systems & Hardware

Optimize single-core top-10 vector retrieval under one- and two-tag predicates

EDA Standard-Cell Gate Sizing Optimization

Systems & Hardware

Minimize leakage while satisfying timing, slew, and capacitance constraints on unseen netlists

EDA Flip-Flop Banking and Placement Optimization

Systems & Hardware

Optimize multi-bit flip-flop grouping and legal placement across power, area, timing, and displacement costs

Quantum Color-Code Decoder

Systems & Hardware

Reimplement a colour-code decoder without the usual libraries

Quantum Circuit TDD Contraction Planning

Systems & Hardware

Choose contraction pairs to minimize peak diagram size and runtime in exact equivalence checking

Ragged GQA Training Kernel Optimization

Systems & Hardware

Accelerate packed variable-length attention with complete Q/K/V backward passes on unseen shapes

Multi-Stream Gated FlashFFTConv Kernel Optimization

Systems & Hardware

Accelerate gated causal long convolutions across independent streams and unseen shapes

Paged Ragged GQA Decode Kernel Optimization

Systems & Hardware

Accelerate grouped-query attention over paged, variable-length KV caches and unseen shapes

TPU v6e Masked GQA Kernel Optimization

Systems & Hardware

Accelerate decode, prefill, and encoder attention across masks, head-sharing ratios, and unseen shapes

Robot Routing and Macro-Program Optimization

Physical Sciences & Engineering

Jointly plan pickup routes and reusable macro recordings to minimize controller button presses

Power-Diode Extreme-Regime I–V Extrapolation

Physical Sciences & Engineering

Predict leakage and conduction currents at destructive voltage-temperature conditions from safe-window measurements

Distribution Feeder Model Calibration from Smart-Meter Data

Physical Sciences & Engineering

Jointly infer customer phases, line impedances, and regulator taps on unseen feeders

Tidal Seabed Friction Field Inversion

Physical Sciences & Engineering

Recover a spatial drag field from sparse, noisy tide-gauge amplitudes and phases on unseen basins

2048 Search-Policy Optimization

Optimization, Planning & Control

Maximize mean game score on unseen seeds under a pooled CPU budget

Query-Budgeted Noisy Black-Box Optimization

Optimization, Planning & Control

Minimize 10-dimensional multimodal objectives with 120 noisy evaluations on unseen instances

Query-Budgeted Stochastic Inventory Policy Optimization

Optimization, Planning & Control

Minimize simulated inventory cost by tuning four replenishment controls within 48 noisy queries

City-Scale Traffic Signal Control Optimization

Optimization, Planning & Control

Coordinate phases, green times, and offsets across ~600 junctions under unseen traffic demand

Fixed-Color-Budget Graph Coloring Optimization

Optimization, Planning & Control

Minimize monochromatic edges on unseen hard graphs under a 45-second solve limit

Defect-Aware Guillotine Glass Cutting Optimization

Optimization, Planning & Control

Pack glass orders into minimum stock plates under staged cuts, defects, and stack-sequence constraints

Cross-Site Chest X-Ray Triage Policy Optimization

Life Sciences & Medicine

Calibrate and rank frozen model scores to prioritize critical studies under a shared review budget

Unsupervised PBMC scATAC Batch Integration

Life Sciences & Medicine

Remove batch effects while preserving cell-type structure in embeddings recomputed on unseen cells

Perturbation-Conditioned Six-Channel Cell Morphology Generation

Life Sciences & Medicine

Generate treated-cell image distributions from controls and condition embeddings without paired examples

Five-Shot Lung scRNA-seq Cell-Type Annotation

Life Sciences & Medicine

Label donor-disjoint cells using five examples per class, an unlabeled pool, and a fixed reference model

Protein–Ligand Co-Folding Inference Optimization

Life Sciences & Medicine

Improve sampling, pose selection, and post-processing under physical-validity and 2 Å RMSD checks

Equity DCF Value-Driver Forecasting

Finance, Law & Business

Forecast growth, margins, reinvestment, WACC, and terminal growth for unseen companies from messy fundamentals

Multi-Matter Legal Research and Analysis Agent

Finance, Law & Business

Answer regulatory and transactional questions across six matters and 1,000+ case documents

Cross-Sectional Stock Return Ranking

Finance, Law & Business

Predict daily forward-return rankings from a 360-factor temporal panel on a sealed later period

Private set

Terminal-Task LLM Agent Harness Design

AI Models & Agents

Redesign the harness around a frozen base model for unseen terminal tasks

Mathematical and Statistical Reasoning Workflow Design

AI Models & Agents

Design one LLM workflow for unseen mathematics and statistics problems

Budget-Constrained 3D Gaussian Splat Compression

AI Models & Agents

Compress a 3D Gaussian scene to a fixed primitive budget while preserving novel-view fidelity

Five-Step Diffusion Sampler Schedule Optimization

AI Models & Agents

Optimize a five-evaluation integration schedule for structured image generation

Budget-Constrained Multi-Agent Medical QA Workflow Design

AI Models & Agents

Design a multi-agent workflow for medical question answering under a fixed model-call budget

Point-Track Candidate Lattice Selection and Fusion

AI Models & Agents

Select and fuse candidate trajectories across a point-track lattice

Query-Conditioned Video Evidence Selection Agent Design

AI Models & Agents

Select temporal evidence from candidate tracklets for vision-language question answering

Ten-Evaluation Diffusion Sampler Optimization

AI Models & Agents

Optimize a generative sampler under a fixed ten-evaluation budget

Label-Free World-Model Hallucination Prediction

AI Models & Agents

Predict imminent world-model hallucinations from unlabeled rollout signals

Nuisance-Robust Semantic Image Editing

AI Models & Agents

Preserve a semantic image edit across unseen nuisance transformations

Cross-Language Dart Formatter Reimplementation

Systems & Hardware

Reimplement the Dart source formatter in Haskell with byte-exact output compatibility

PostgreSQL-Compatible Server over SQLite

Systems & Hardware

Implement the PostgreSQL wire protocol in Zig over a SQLite storage engine

EDA Cell Placement, Sizing, and Buffering Optimization

Systems & Hardware

Refine an existing placed netlist to improve power, performance, and area

Mamba-3 Training Kernel Optimization

Systems & Hardware

Optimize Mamba-3 scans with joint states, long ragged sequences, and a full backward pass

Building Indoor-Temperature Extreme Extrapolation

Physical Sciences & Engineering

Predict summer indoor temperatures from a single mild-season building record

Climate Subgrid Convection Emulation

Physical Sciences & Engineering

Learn a neural parameterization of unresolved convection for a climate model

Dual-Doppler 3D Wind-Field Inversion

Physical Sciences & Engineering

Reconstruct three-dimensional wind fields from dual-Doppler radar observations when vertical motion is underdetermined

Online Hybrid-Vehicle Energy Management Policy Optimization

Physical Sciences & Engineering

Optimize causal power allocation using only information available to the vehicle online

Exoplanet Transmission-Spectrum Atmospheric Retrieval

Physical Sciences & Engineering

Infer atmospheric properties from unseen exoplanet transmission spectra

Sparse-Sensor Damage Mapping

Physical Sciences & Engineering

Localize stiffness loss in a steel structure

Fuel-Efficient Real-Time Gear-Shift Policy Optimization

Physical Sciences & Engineering

Optimize second-by-second gear selection to reduce fuel use on unseen driving cycles

Molecular HOMO–LUMO Gap Prediction from SMILES

Physical Sciences & Engineering

Predict orbital energy gaps for unseen molecules from their SMILES strings

Constrained Ackermann Parking Trajectory Optimization

Physical Sciences & Engineering

Plan collision-free parking trajectories for an Ackermann-steered vehicle in tight slots

Photometric Transient and Variable-Source Classification

Physical Sciences & Engineering

Classify astronomical transients and variable sources from multiband light curves

Time-Optimal Drone Racing

Physical Sciences & Engineering

Fly a quadrotor through gates in minimum time

Suspension Geometry Co-Design

Physical Sciences & Engineering

Place suspension hardpoints and pick the tuning that goes with them

Extrapolative Scientific Law Discovery

Physical Sciences & Engineering

Discover equation families that extrapolate beyond the observed data rather than only fitting it

Borehole Lithology Sequence Classification

Physical Sciences & Engineering

Classify lithology facies along unseen boreholes from multichannel wireline logs

Wake-Aware Wind Farm Layout Optimization

Physical Sciences & Engineering

Place turbines to minimize wake losses and maximize energy yield on unseen sites

Query-Budgeted Continuous Black-Box Optimization

Optimization, Planning & Control

Design a self-contained optimizer for unseen 10-dimensional objectives under a 100-evaluation budget

Bilevel Road Capacity and Toll Design

Optimization, Planning & Control

Choose capacity expansions and tolls while accounting for the resulting traffic equilibrium

Self-Financing Rail Network Expansion Planning

Optimization, Planning & Control

Stage rail expansions so operating revenue funds subsequent construction

Time-Windowed Robotaxi Dispatch Optimization

Optimization, Planning & Control

Dispatch a robotaxi fleet for paired pickup and drop-off requests under two-sided time windows

Electric-Van Routing with Charging Detours

Optimization, Planning & Control

Route electric vans while jointly selecting charging detours under battery constraints

Long-Horizon Transport Company Planning Agent

Optimization, Planning & Control

Manage an OpenTTD transport company over a 20-year simulated horizon

Recursive Sokoban Planning Agent

Optimization, Planning & Control

Solve Sokoban instances whose rooms recursively contain other rooms

Large-Scale Pickup-and-Delivery Routing with Time Windows

Optimization, Planning & Control

Route vehicles to serve 500 paired pickup and delivery requests within their time windows

Online Railway Disruption Recovery and Rescheduling

Optimization, Planning & Control

Reschedule trains under unforeseen breakdowns and evolving network availability

Collision-Free Multi-Crane Scheduling

Optimization, Planning & Control

Schedule five terminal cranes that share a rail and cannot pass one another

Real-Time Battle Tetris Planning Agent

Optimization, Planning & Control

Design an agent that survives holds, combos, and incoming garbage under real-time decision constraints

Coverage-Constrained Urban Bus Network Design

Optimization, Planning & Control

Design a bus route network that serves every required node

DNA-to-ATAC Accessibility Prediction

Life Sciences & Medicine

Predict base-resolution chromatin accessibility directly from genomic DNA sequence

12-Lead ECG Chagas Disease Screening

Life Sciences & Medicine

Detect likely Chagas disease cases from unseen 12-lead ECG recordings

Chromatin-to-Gene Expression Prediction

Life Sciences & Medicine

Predict single-cell gene expression from chromatin accessibility profiles alone

Mini-Protein Stability Regression

Life Sciences & Medicine

Train and package a model that predicts stability for unseen mini-protein sequences

Out-of-Distribution Cell Morphology Generation

Life Sciences & Medicine

Generate perturbed-cell morphologies that preserve the requested mechanism of action under distribution shift

Cross-Dataset Single-Cell Label Transfer

Life Sciences & Medicine

Annotate cells across unseen datasets and protocols with a frozen foundation model

Zero-Shot TCR–Epitope Binding Prediction

Life Sciences & Medicine

Score TCR binding to epitopes absent from the training set

Investment Banking Deal-Document Question Answering

Finance, Law & Business

Answer transaction and diligence questions grounded in underlying deal files

Management Consulting Engagement-Document Question Answering

Finance, Law & Business

Answer consulting questions grounded in underlying engagement files

Arbitrage-Free Implied-Volatility Surface Calibration

Finance, Law & Business

Calibrate an arbitrage-free volatility surface from a single noisy market snapshot

Transactional Legal Matter Document Reasoning

Finance, Law & Business

Resolve transactional legal matters across a 745-document case corpus

Multivariate Time-Series Structural Break Detection

Finance, Law & Business

Detect regime changes across a fleet of noisy time series