TAKO demonstrates real-time adversarial takeover of robotic diffusion policies via reusable universal patches on visual inputs, achieving 100% success in steering attacker-chosen trajectories across multiple tasks, encoders, and diffusion methods.
hub Canonical reference
K., Garcia Lambas, D., Ruiz, A
Canonical reference. 73% of citing Pith papers cite this work as background.
31
Pith papers citing it
Background
73% of classified citations
hub tools
citation-role summary
background 10
method 1
citation-polarity summary
representative citing papers
IVE generates a linearly independent vector from powers of an encrypted index in O(p) FHE operations instead of O(p log p) for one-hot, then recovers the embedding via orthogonal DCT change of basis, yielding up to 78.4x faster amortized lookup and reducing vector generation share from 99.6% to 66.3
MMM packages knowledge as typed vertices, edges and pens with free-text labels under a small set of normative rules, enabling structural interoperability and decentralisation without semantic convergence.
Signature filtering learns unreliable tokens with MILP and removes them at detection time, raising true positive rates from 8-31% to 78-99% across Kgw, Sweet, Unigram, and Exp watermarks on multiple corpora and LLMs while controlling false positives.
SuperMemory-VQA provides 4,853 human-verified QA pairs from 52.9 hours of egocentric AI glasses recordings to benchmark AI systems on realistic long-horizon memory tasks including an unanswerable option.
LaneRoPE adds an inter-sequence attention mask and extended RoPE to enable collaborative parallel sequence generation in LLMs, yielding accuracy gains on math reasoning under length limits.
JobBench is a new benchmark with 130 occupational tasks where the best of 36 tested AI models achieves only 45.9% success.
CodeQL detected 171 CVEs total, with 83 caught by a prior version before the fix; detections were often actionable within the vulnerable file but not stable across tool versions.
A new structured prompting method (SPEC) helps AI detect insufficient evidence in adjudication tasks and defer decisions appropriately, reaching 89% accuracy on a benchmark varying information completeness from Colorado unemployment insurance cases.
CoDiLA adds a compact auxiliary AR model on diffusion latents to enforce local sequential validity during parallel token sampling in discrete diffusion language models.
A calibrated thick-disk semianalytic model plus MCMC retrieval quantifies SED constraints on CPD parameters and recovers consistent masses and accretion rates for PDS 70 b/c and GQ Lup b.
PACE selects execution horizons online via low-speed valleys in predicted action chunks, boosting task success by 6.4 points in simulation and 19.7 points on real robots.
Engagement Process (EP) decouples actions and observations as independent event streams over time within a POMDP structure to explicitly model temporal dynamics in agent interactions.
AI peer review systems are vulnerable to prompt injections, prestige biases, assertion strength effects, and contextual poisoning, as demonstrated by a new attack taxonomy and causal experiments on real conference submissions.
FASTER models multi-candidate denoising as an MDP and trains a value function to filter actions early, delivering the performance of full sampling at lower cost in diffusion RL policies.
FRB dispersion measures directly constrain suppression of the matter power spectrum due to feedback at k ~ 0.1-3 h/Mpc, reduce posterior variance by a factor of ~8 at k~1 h/Mpc, and exclude extreme large-scale feedback scenarios at ~2 sigma.
MAESTRO adds a shared preference memory plus GUI-adaptation and workflow-navigation mechanisms to conversational agents with GUIs and tests them in a 33-person movie-booking study.
GWTC-4 data analysis yields a pair-instability mass gap lower edge at 44.3^{+5.9}_{-3.5} M_⊙, an S-factor of 268^{+195}_{-116} keV b for ^{12}C(α,γ)^{16}O, and two populations supporting both direct formation and hierarchical mergers.
SKA FRB dispersion measures and their cross-correlations with Stage IV shear and clustering can pin down baryonic feedback and improve cosmological constraints by factors of ~2–5 under optimistic detection rates.
Coarse-to-Control adds planning via coarse action tokens in the same vocabulary as control actions, improving VLA performance on long-horizon manipulation tasks.
Void x CMB lensing from Roman mocks is robust to catalog construction choices and forecasts S/N of 13-31 sigma with Planck, SO, and CMB-S4-like data for 2D and 3D voids.
Derives an asymptotic equivalent for the Representation Gap in equivariant diffusion models, showing it depends primarily on the intrinsic dimension of the task.
Function-based chunking underperforms other strategies in RAG code completion by 3.57-5.64 points, with context length as the dominant factor.
The study uncovers symmetry-protected topological phases with coexisting spontaneous symmetry breaking in the triangular Majorana-Hubbard ladder model of interacting Majorana fermions.
citing papers explorer
-
Test-time Adversarial Takeover: A Real-time Hijacking Interface against Robotic Diffusion Policies
TAKO demonstrates real-time adversarial takeover of robotic diffusion policies via reusable universal patches on visual inputs, achieving 100% success in steering attacker-chosen trajectories across multiple tasks, encoders, and diffusion methods.
-
Private Embedding Lookup with Encrypted Compact Queries under Fully Homomorphic Encryption
IVE generates a linearly independent vector from powers of an encrypted index in O(p) FHE operations instead of O(p log p) for one-hot, then recovers the embedding via orthogonal DCT change of basis, yielding up to 78.4x faster amortized lookup and reducing vector generation share from 99.6% to 66.3
-
The MMM Data Model -- A Normative Specification for Knowledge Interoperability in a Decentralisable Knowledge Commons
MMM packages knowledge as typed vertices, edges and pens with free-text labels under a small set of normative rules, enabling structural interoperability and decentralisation without semantic convergence.
-
Signature filtering: a lightweight enhancement for statistical watermark detection in large language models
Signature filtering learns unreliable tokens with MILP and removes them at detection time, raising true positive rates from 8-31% to 78-99% across Kgw, Sweet, Unigram, and Exp watermarks on multiple corpora and LLMs while controlling false positives.
-
SuperMemory-VQA: An Egocentric Visual Question-Answering Benchmark for Long-Horizon Memory
SuperMemory-VQA provides 4,853 human-verified QA pairs from 52.9 hours of egocentric AI glasses recordings to benchmark AI systems on realistic long-horizon memory tasks including an unanswerable option.
-
LaneRoPE: Positional Encoding for Collaborative Parallel Reasoning and Generation
LaneRoPE adds an inter-sequence attention mask and extended RoPE to enable collaborative parallel sequence generation in LLMs, yielding accuracy gains on math reasoning under length limits.
-
JobBench: Aligning Agent Work With Human Will
JobBench is a new benchmark with 130 occupational tasks where the best of 36 tested AI models achieves only 45.9% success.
-
Longitudinal Analyses of SAST Tools: A CodeQL Case Study
CodeQL detected 171 CVEs total, with 83 caught by a prior version before the fix; detections were often actionable within the vulnerable file but not stable across tool versions.
-
Learning When Not to Decide: A Framework for Overcoming Factual Presumptuousness in AI Adjudication
A new structured prompting method (SPEC) helps AI detect insufficient evidence in adjudication tasks and defer decisions appropriately, reaching 89% accuracy on a benchmark varying information completeness from Colorado unemployment insurance cases.
-
Locally Coherent Parallel Decoding in Diffusion Language Models
CoDiLA adds a compact auxiliary AR model on diffusion latents to enforce local sequential validity during parallel token sampling in discrete diffusion language models.
-
A Retrieval Framework for Observationally Constraining the Parameters of Circumplanetary Disks
A calibrated thick-disk semianalytic model plus MCMC retrieval quantifies SED constraints on CPD parameters and recovers consistent masses and accretion rates for PDS 70 b/c and GQ Lup b.
-
PACE: Phase-Aware Chunk Execution for Robot Policies with Action Chunking
PACE selects execution horizons online via low-speed valleys in predicted action chunks, boosting task success by 6.4 points in simulation and 19.7 points on real robots.
-
Engagement Process: Rethinking the Temporal Interface of Action and Observation
Engagement Process (EP) decouples actions and observations as independent event streams over time within a POMDP structure to explicitly model temporal dynamics in agent interactions.
-
When AI reviews science: Can we trust the referee?
AI peer review systems are vulnerable to prompt injections, prestige biases, assertion strength effects, and contextual poisoning, as demonstrated by a new attack taxonomy and causal experiments on real conference submissions.
-
FASTER: Value-Guided Sampling for Fast RL
FASTER models multi-candidate denoising as an MDP and trains a value function to filter actions early, delivering the performance of full sampling at lower cost in diffusion RL policies.
-
Signatures of Suppressed Matter Clustering revealed by Fast Radio Bursts
FRB dispersion measures directly constrain suppression of the matter power spectrum due to feedback at k ~ 0.1-3 h/Mpc, reduce posterior variance by a factor of ~8 at k~1 h/Mpc, and exclude extreme large-scale feedback scenarios at ~2 sigma.
-
MAESTRO: Adapting GUIs and Guiding Navigation with User Preferences in Conversational Agents with GUIs
MAESTRO adds a shared preference memory plus GUI-adaptation and workflow-navigation mechanisms to conversational agents with GUIs and tests them in a 33-person movie-booking study.
-
Gravitational-wave constraints on the pair-instability mass gap and nuclear burning in massive stars
GWTC-4 data analysis yields a pair-instability mass gap lower edge at 44.3^{+5.9}_{-3.5} M_⊙, an S-factor of 268^{+195}_{-116} keV b for ^{12}C(α,γ)^{16}O, and two populations supporting both direct formation and hierarchical mergers.
-
Probing the Baryon Distribution with Fast Radio Bursts
SKA FRB dispersion measures and their cross-correlations with Stage IV shear and clustering can pin down baryonic feedback and improve cosmological constraints by factors of ~2–5 under optimistic detection rates.
-
Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models
Coarse-to-Control adds planning via coarse action tokens in the same vocabulary as control actions, improving VLA performance on long-horizon manipulation tasks.
-
Towards precision cosmology with Void x CMB correlations (II): Impact of mock catalogs on the Void x CMB lensing signal
Void x CMB lensing from Roman mocks is robust to catalog construction choices and forecasts S/N of 13-31 sigma with Planck, SO, and CMB-S4-like data for 2D and 3D voids.
-
Representation Gap: Explaining the Unreasonable Effectiveness of Neural Networks from a Geometric Perspective
Derives an asymptotic equivalent for the Representation Gap in equivariant diffusion models, showing it depends primarily on the intrinsic dimension of the task.
-
How Does Chunking Affect Retrieval-Augmented Code Completion? A Controlled Empirical Study
Function-based chunking underperforms other strategies in RAG code completion by 3.57-5.64 points, with context length as the dominant factor.
-
Symmetry-Protected Topological Phases in the Triangular Majorana-Hubbard Ladder
The study uncovers symmetry-protected topological phases with coexisting spontaneous symmetry breaking in the triangular Majorana-Hubbard ladder model of interacting Majorana fermions.
-
Confidence Without Competence in AI-Assisted Knowledge Work
Standard LLM chats produce high perceived understanding but low objective learning in students, while future-self explanations best align confidence with actual gains and guided hints maximize learning with moderate workload.
-
A Machine Learning Framework for Turbofan Health Estimation via Inverse Problem Formulation
A new turbofan dataset with realistic maintenance patterns is used to benchmark Bayesian filters as strong baselines against self-supervised learning representations for component health estimation.
-
Symmetry-guided and AI-accelerated design of intercalated transition metal dichalcogenides for antiferromagnetic spintronics
Symmetry-guided graph neural networks trained on 200 structures screen 100,000+ iTMD configurations to identify 55 antiferromagnetic spintronics candidates with d-wave altermagnetism or giant T-odd spin Edelstein effects.
-
Unveiling the roles of thermal and nonthermal processes in the ISM & IGM structure formation and evolution of galaxies with SKAO
Simulations indicate SKAO AA4 surveys can trace thermal and nonthermal ISM processes in high-redshift galaxy analogs beyond z=1-3, underscoring nonthermal feedback at cosmic noon.
-
Constraints on axion-like particles from ultra-high-energy observations of M87 with the HAWC observatory
No photon-ALP conversion signal found in HAWC data from M87, producing competitive constraints excluding ALP masses of 10^{-8} to 10^{-6} eV for couplings above 5×10^{-12} GeV^{-1}.
-
Prompt Governance? On Governing Technologies Governed by Natural Language
Literature on system prompts for AI shows fragmented and contradictory claims that complicate policy efforts to use them as reliable governance mechanisms.
-
Multi-Agent Consensus as a Cognitive Bias Trigger in Human-AI Interaction
Majority consensus among AI agents speeds up human opinion change and raises confidence via social proof, while minority dissent slows it and encourages more deliberation, based on an experiment comparing three agent configurations.