Skip to main content
Search

KTH Royal Institute of Technology

12 related articles

Subscribe to updates
  1. arXiv

    Coding agents must now declare why before every tool call: on deliberately broken tests, the journal exposed 90% of cheating runs

    Coding agents change code through tool calls, but a call carries no intent, and permission rules, logs and LLM monitors see only command text or after-the-fact effects. Episteme makes a type-plus-reason intent declaration mandatory for every tool call, with the harness recording actual effects and checking them against the declaration. On ImpossibleBench, whose tasks can be passed only by cheating, the journal exposed 44 of 49 misbehaving runs across two open-weight models, a 90% exposure rate, all through the agents' own stated reasons.

  2. arXiv

    Traffic forecasting drops step-by-step sampling: CoReFlow emits a full forecast in one pass and wins 11 of 12 point-metric cases across four road networks

    Probabilistic traffic forecasting has relied on multi-step diffusion sampling, which is slow and often trails deterministic baselines on point accuracy. CoReFlow learns the flow map, the solution operator of the probability-flow ODE, directly and relaxes trajectory consistency through the conditional mean, so a single forward pass produces the full forecast. On four real road-network datasets, PEMS03, PEMS04, PEMS07 and PEMS08, it ranks first in 11 of 12 deterministic metric cases and takes all 14 reported comparisons against generative baselines on PEMS07 and PEMS08.

  3. bioRxiv

    Neighbor transcripts leak into spatial cells: DisCell splits spillover from niche response and gives each finding a breakdown point

    Single-cell spatial transcriptomics counts neighboring cells' transcripts inside a cell, and this spillover and a cell's real response to its niche both tie expression to surroundings, so counts alone cannot separate them. DisCell models intrinsic state, niche response and spillover jointly, refits over a range of spillover fractions, and assigns each finding a breakdown point. On ovarian and lung sections its intrinsic state keeps more cell-cycle information than any competing method holding as little neighborhood information, and a published esophageal EMT finding survives the modeled spillover range.

  4. Cyber defense simulators stop hard-coding the network: swap the modeling language, train attackers and defenders in one environment

    The KTH team released the MAL Simulator, which builds cyber attack and defense simulation on the Meta Attack Language so a new target domain can be modeled without changing simulator source code. Across 30 scenarios derived from the ADS-24 network in the CRATE cyber range, a trained reinforcement-learning attacker returned −301±10.3 on 15 unseen test scenarios, better than breadth-first, depth-first and random search, and a defender trained under 1% false and missed alerts returned −60355±945 against −61653±914 for a heuristic defender. Against the reinforcement-learning attacker, every defender's return dropped markedly.

  5. arXiv

    Human motion generation goes image diffusion: joint trajectories become images, and two skeletons share one model

    Text-to-motion generation no longer needs a separate model per skeleton: once root-centered joint trajectories are arranged into joint-time images, image diffusion architectures share one motion autoencoder and one generator across HumanML3D and KIT-ML while keeping each dataset's native skeletal definition.

  6. arXiv

    Controller retuning now screens stability first: compressor simulation deploys zero unstable candidates

    On an axial compressor-plenum-throttle simulation, UQ-ACE infers a posterior over local dynamics from closed-loop data and blocks candidates whose stability probability falls short, producing no unstable trajectories across 100 runs.

  7. Live-cell nanoscopy stops burning out: tenfold less light, fourfold faster frames, and mitochondria caught budding

    Two deep-learning schemes extend RESOLFT live-cell nanoscopy to roughly five times longer recordings and four times higher frame rates, capturing 80 to 200 nanometer mitochondrial protrusions at 2 hertz.

  8. medRxiv

    Coronary risk scores stop giving one number: eight biological axes separate people with the same genetic liability

    Polygenic risk for coronary artery disease no longer collapses into a single number. Conventional scores identify who is at high risk but not which biological processes drive that risk. Decomposing 526 genome-wide significant variants by their associations with 78 CAD-related traits yielded eight latent components: cholesterol-enriched lipoproteins, triglyceride-rich lipoproteins, lipoprotein(a), metabolically unhealthy normal weight, obesity, endothelial dysfunction, coagulation and cardiac remodeling. Process-specific scores built in UK Biobank, SCAPIS and Mass General Brigham Biobank tracked clinical coronary events, subclinical atherosclerosis and distinct biomarker and comorbidity profiles.

  9. arXiv

    ICDP constrains interaction distribution shift via contrastive density-ratio estimation, suppressing high-value but interaction-unsupported trajectories on nuPlan closed-loop and a real truck

    The work introduces ICDP, an offline reinforcement learning framework that exactly decomposes joint-support degradation over ego and surrounding-agent futures into an ego-support component and a residual interaction-support component, recovers the latter through contrastive density-ratio estimation, and constrains policy improvement without explicit joint-density modeling, surrounding-agent prediction, or reactive simulator or world-model rollouts; on nuPlan and InterPlan closed-loop evaluations and real-world truck experiments, ICDP suppresses high-value yet interaction-unsupported trajectory selections and improves performance in interaction-critical driving scenarios.

  10. arXiv

    CIPO trains LLM agents with counterfactual branches to adaptively pick tool granularity, raising success and decision compression across three benchmarks and two backbones

    The work proposes CIPO: it first mines executable composite skills from successful tool-use trajectories via budget-constrained BPE-style merging, then augments GRPO by branching at the first eligible granularity decision of each base rollout and using the paired outcome difference as a bounded supplementary reward; across TOOLATHLON, TOUCAN, and TRAJECT-Bench with Qwen2.5-7B and Llama3.1-8B, CIPO attains the highest Tool F1 and TSR in all six settings with an average Comp./Raw of 0.845 versus 0.912 for GRPO + Skills, and the gain does not come from simply invoking skills more often.