// intelligence · research
AI Research Papers
Preprints and publications from arXiv, Berkeley AI Research, MIT, Google Research, and leading academic institutions worldwide.
500 papers · showing 20
Introducing Candela: AI risk benchmarking-as-a-service for developers and deployers
Benchmark results should be believable, useful, and comparable. Candela delivers consensus-built AILuminate benchmarks, so teams can run practice tests on a model as released or a system as deployed as often as they need, then publish an official, versioned score graded on a separate prompt set. The post Introducing Candela: AI risk benchmarking-as-a-service for developers and deployers appeared first on MLCommons.
Test Your Own AI Against Real-World Risks and Attacks with AILuminate Safety 1.2.1 and Jailbreak 1.1
New versions of the AILuminate Safety and Jailbreak benchmarks are available in the new Candela benchmarking-as-a-service product, and feature a custom evaluator, percentile scoring, and resistance to benchmaxxing, cherrypicking, and other forms of gaming. Safety scores are live, Jailbreak scores coming soon. The post Test Your Own AI Against Real-World Risks and Attacks with AILuminate Safety 1.2.1 and Jailbreak 1.1 appeared first on MLCommons.
Don’t assume composability in AI safety. Deployments need their own testing.
Teams deploy whole systems but cite safety scores from base models. Fine-tunes drift safety in unpredictable directions, and many benchmarks are too noisy to detect the change. This post explains why safety scoring is not composable, and what MLCommons is doing to make it comparable. The post Don’t assume composability in AI safety. Deployments need their own testing. appeared first on MLCommons.
Crunching numbers behind closed doors
A secure data analytics platform enables cross-institutional studies of patient health records without pooling raw data or compromising privacy.
Getting hired in AI: Navigating Canada’s evolving AI job market
Canadian AI hiring is shifting from generalist credentials to specialized technical execution. In Ontario alone, more than 71,000 professionals required AI skills this year, representing a massive 61 per cent […] The post Getting hired in AI: Navigating Canada’s evolving AI job market appeared first on Vector Institute for Artificial Intelligence.
3 Questions: Why infrastructure is entering a defining era
Professor Ali Jadbabaie describes how AI, digital transformation, climate change, and complex geopolitics are forcing engineers and policymakers to rethink how infrastructure is designed, built, and managed.
CIFAR welcomes four new members to its Board of Directors
Cherie Brant, Claude Généreux, Paul Sabourin and Annette Trimbee bring extensive experience spanning business, finance, higher education, public service and community leadership. CIFAR is pleased […] The post CIFAR welcomes four new members to its Board of Directors appeared first on CIFAR.
Closed-Loop Aware Planning via Imitation-Free Speed Scoring for End-to-End Autonomous Driving
arXiv:2610.36079v1 Announce Type: new Abstract: End-to-end autonomous driving has predominantly relied on open-loop imitation learning for training, which presents two limitations. First, expert demonstrations show only one behavior per scenario, so the model cannot learn alternative valid behaviors and, in turn, cannot adapt to user-specified preferences at test time. Second, noise or ambiguity in these demonstrations can produce inconsistent, unreliable supervision for the end-to-end models. Driving is inherently multi-modal: multiple different behaviors can be valid responses to the same scene, and this diversity arises largely from speed rather than path. We introduce a planning paradigm that learns the path through imitation while scoring a set of candidate speeds across multiple criteria with a learned function supervised by physically grounded safety constraints. We realize this paradigm as Closed-Loop Aware Planning for Autonomous Driving (CLAP-AD), a plug-in planning head that integrates with existing end-to-end driving models and, to our knowledge, is the first to enable explicit test-time control over their speed selection without speed-conditioned training. On Bench2Drive-Speed, CLAP-AD sets a new state of the art Speed-Adherence Score of 82.53, topping TFv6, the strongest baseline on this metric, by 12.9 points while retaining strong closed-loop driving performance. On Bench2Drive, it improves driving score by +28.5 and +32.7 over UniAD and VAD, and on NAVSIM v2, built on real-world driving data, EPDMS by +6.1 over its backbone. Project page: https://clap-ad.github.io
Algorithmics for Graphical Separation Criteria
arXiv:2610.36502v1 Announce Type: new Abstract: Statistical graphical models are representations of the dependences between the components of a random vector. They encode these dependences via graphical separation criteria. We consider the algorithmic problem of checking separation statements for the various graph classes and separation criteria that have been developed by the community. We establish that all criteria we study can be checked by modular variations of reachability. For all separation criteria but $\sigma$-separation, this yields a linear-time algorithm. This is effectively the optimal attainable time since loading the graph itself is linear-time in the number of edges in the graph. Using the CIfly framework for deploying modular variations of reachability, we also provide R and Python code implementing our algorithms. We evaluate empirical runtime and scaling behavior in a simulation study.
Learning Human Sports Motion Feasibility via Quantile Regression on Tracking Data
arXiv:2610.36183v1 Announce Type: new Abstract: We study time-to-arrival (TTA) in sports: the shortest plausible time a player needs to reach an arbitrary location on the pitch given their current kinematic state, estimated as the lower envelope of observed locomotion times. Rather than modeling space ownership through position density estimation or closed-form physics-based models, we cast TTA as a conditional quantile regression problem, estimating the lower quantile of locomotion time so the model learns a player's lower-envelope arrival time. We train the model on optical tracking data from all 64 matches of the PFF FC FIFA World Cup 2022 dataset, and evaluate a lightweight multilayer perceptron against six player-motion models from the literature, re-expressed as TTA predictors, and a learned reachable- hull baseline. Across our evaluation metrics, the MLP consistently outperforms every physics-based baseline. To probe the generality of the learned locomotion model, we further evaluate it on basketball tracking data and a single-game NFL tracking dataset, analyzing the transferability of player motion dynamics across sports. Our project is available at https://github.com/mguti97/TTA-Quantile.
Lessons from Luck: Do LLM-Written Lessons Carry Chance into an Agent's Next Decision?
arXiv:2610.33029v1 Announce Type: new Abstract: Self-improving agents write lessons from the outcomes of their runs; when outcomes are noisy, a lesson can carry chance into later decisions. We measure one link of this chain in a registered study. In 256 synthetic worlds, each a recurring task with two interchangeable tools of recorded reliability, a scripted previous run used the more reliable tool (a good decision) or the less reliable one (a bad decision), with an assigned outcome: success or a transient failure. A model wrote a lesson from the run with ReasoningBank's memory-writing instruction; a fresh session of the same model, holding the authoritative log and the previous result, chose the next run's first tool with one memory item: that lesson, the lesson from the other outcome, an irrelevant item, or an unreasoned directive. The protocol was frozen, deposited and timestamped before any confirmatory call; 11 LLM endpoints gave 90,084 answers. After a chance failure of the better tool, the panel's lessons carried the chance (registered reading luck-carrying; interaction +0.112, 95% interval [+0.097, +0.126]): readers named the worse tool in 11.6% of next runs with the lesson written from that failure, against 0.4% with an irrelevant item and 0.4% with the lesson written from a success. The effect was +0.042 where the log stated that failures are independent and +0.181 where it did not (descriptive). After a lucky success of the worse tool, the interaction was bounded below 0.10 (+0.002), a bound, not evidence of no harm. In exploratory readings, 6 endpoints carried the chance (up to +0.352); 5 were bounded below 0.10. In secondary comparisons, a three-sentence decision-quality note in the writing instruction lowered the worse-tool rate after a chance failure to 7.5%, and one instruction for both outcomes with a no-change option to 3.1%; neither met the registered harm-reduction rule at every such endpoint.
Who Convinced Whom? Tracing How LLMs Change Their Minds in Multi-Agent Systems
arXiv:2610.35930v1 Announce Type: new Abstract: Multi-agent debate (MAD) aims to improve language-model decisions through the exchange of arguments, yet agreement after debate need not reflect independent corroboration. We investigate how peer communication influences individual decisions and collective outcomes through behavioral and internal interventions. We introduce MAD Bench, a 2,400-question benchmark spanning knowledge and reasoning, values and moral judgment, and safety and refusal, and evaluate three open-weight models using a round-matched independent-reflection control. Among answer-changing responses, a forced-choice readout before rationale generation matches the eventual answer more often in debate than in reflection (65.4% versus 22.3% on Qwen3.5-9B). Message ablations reveal model-dependent effects of peer answers and reasoning, while attributing identical text to the receiver rather than another agent substantially changes its influence. For the tested source descriptions, stated track record affects answer preferences more strongly than professional-status cues across all three models. Attention masking produces different changes in decision margins across the three checkpoints. On fully parsed, reference-labelled non-numeric questions, debate yields small estimated accuracy improvements over initial majority voting, while increasing the rate of unanimous wrong answers by 3.1-7.4 percentage points relative to reflection. Longer debates show persistent consensus and greater correlation between agents' correctness than reflection. These findings show that stronger consensus can coexist with more shared errors, motivating evaluations that assess dependence and unanimous errors alongside accuracy.
A functional model for hyponormal contractions
arXiv:2610.35880v1 Announce Type: new Abstract: In this paper, we establish an analytical model for a class of hyponormal contractions $T$ satisfying $Ran[T^*,T] \subseteq \mathcal D_{T^*}\ominus \mathcal D_T$, where $\mathcal D_{T^*}$ and $\mathcal D_T$ are the associated defect spaces for $T$. We also show that, under a certain condition, a contraction $T$ with $\mathcal D_T \subseteq \mathcal D_{T^*}$ is hyponormal.
Anytime Valid Confidence Bands for Distribution Functions
arXiv:2610.36149v1 Announce Type: new Abstract: Nonparametric inference often requires characterizing an unknown function, such as a cumulative distribution function (CDF), across its entire domain rather than a specific point. We develop confidence bands for distribution functions of a data stream that are valid simultaneously over all times and all points, even when each observation may depend arbitrarily on the past. We begin by deriving such bands for the time-averaged conditional CDF and show how their widths are determined by the depth of a binary search tree. We then show that this result is an instance of a framework through which one can construct valid confidence bands uniformly over time and over an arbitrary index set for any indexed family of time-averaged distribution functions. We illustrate the framework on importance-weighted and right-censored CDFs, as well as multivariate CDFs.
Second-order Sinkhorn Geometry and $\sqrt{n}$-Inference
arXiv:2610.36488v1 Announce Type: new Abstract: The Sinkhorn divergence, a debiased form of entropic optimal transport, has been widely used in various scientific disciplines due to its computational and statistical tractability. Unlike the Wasserstein distance, however, the Sinkhorn divergence is not a metric and it does not naturally admit a formal Riemannian structure. These properties are central to many applications of optimal transport methods that utilize interpolation, barycentric averaging, parallel transport or other geometric operations. Recent work addresses these shortcomings by constructing a formal Riemannian tangent space and a metric tensor on the space of probability measures that quadratically approximates the Sinkhorn divergence. In this work, we extend this framework with two major threads of contributions. For a suitable class of finite-smoothness cost functions, we prove Hadamard differentiability of the Sinkhorn metric tensor and we derive the Levi-Civita covariant derivative in this geometry; we then use the covariant derivative to obtain the geodesic and parallel transport equations, each of which amount to an ODE on a fixed Reproducing Kernel Hilbert Space (RKHS) with smooth coefficients. Our second contribution is to linearize the solutions to the geodesic and parallel transport equations using standard tools from Banach space ODE theory. We show that under standard regularity conditions these linearizations imply $\sqrt{n}$-limiting Gaussian processes centered at population values for the plugin estimators of the exponential map, the logarithmic map, parallel transport and the induced Sinkhorn distance. Moreover, we show that the nonparametric bootstrap is consistent for all objects mentioned. To the best of our knowledge, these are the first $\sqrt{n}$--limit theorems and bootstrap consistency results for canonical geometric operations in a transport-based geometry on the space of probability measures.
Analog Quantum Experiments Meet Automatic Differentiation - from Programs to Atoms
arXiv:2610.36167v1 Announce Type: new Abstract: Gradient-based optimization of analog quantum hardware poses a concrete programming problem: given an executable experiment P(\theta) that returns only measurement samples, what executable experiment should the system run to obtain (d/d\theta_ell)P(\theta)? Conventional automatic differentiation cannot propagate through the inaccessible quantum state, while finite differences require a noise-sensitive step size and analytic gradient identities cover only parts of an experiment. We present DiffSimuQ, which shows that a broad class of open-loop analog quantum experiments is closed under automatic differentiation. A compositional transformation maps P(\theta) to a derivative program D_ell[P(\theta)] in the same language whose expected outcome is exactly the derivative. Only continuous evolution requires a strategy-specific rule, and we give two realizations. Our Nyquist shift rule (NSR) applies to general time-dependent evolution: we prove that the observable response to shifting the Hamiltonian along its tangent is band-limited even for non-commuting controls, enabling exact reconstruction from shifted executions. For involutory generators, a parameter-shift realization (PSR) inserts a parameter-independent operation at a sampled time. Because derivative programs remain ordinary programs, source and derivative experiments share one compiler, which reuses the source realization, lowers branches to device-ready neutral-atom controls, and selects between NSR and PSR based on estimated cost. Under device noise, the exact derivative programs differentiate the objective realized by the device, with hardware realization errors treated separately. In machine-level noisy emulation, NSR and PSR avoid the finite-difference error floor and improve closed-loop optimization; they win in complementary cost regimes, while each derivative branch at 1000 qubits add at most 0.3% to source compilation time.
Accelerating fits in high-energy physics and beyond
arXiv:2610.35927v1 Announce Type: new Abstract: Data analysis in particle physics often involves optimizing a loss function that quantifies how well a theoretical model describes the observed data, in order to estimate the parameters of the model. This is referred to as fitting. As datasets grow larger and more expensive analysis methods are used, the time required to perform this parameter estimation can be a limiting factor. This work explores techniques that can significantly accelerate the fitting process, and measures their performance on a set of example models chosen to be representative of a variety of physics fits. We compare the speed and precision of variants of several well-known optimization algorithms, namely Newton's method, BFGS, and the Migrad algorithm (commonly used in physics fits). We observe significant speedups when using some algorithms: most notably, some variants of BFGS are consistently about $10\times$ faster than Migrad across our benchmarks, while achieving the same precision. For quasi-Newtonian algorithms like BFGS and Migrad, we find that the initial value of the Hessian matrix has a large impact on the runtime. We additionally discuss methods for accelerating the evaluation of the loss function and its derivatives, using algebraic simplifications, automatic differentiation, and modern hardware features on the CPU and GPU. We find that basic algebraic simplifications of the loss function can significantly improve speed and precision. On the GPU we explore using mixed precision computation, and show that this achieves much higher throughput on most accelerators without significant precision loss. Lastly we discuss efficient parallelization strategies for both the CPU and the GPU. All of the techniques and algorithms in this paper are implemented in our new FastFit library, which was used for all of our benchmarks.
SG^2SLAM: Spatial Geometry-Informed 3D Gaussian Splatting SLAM for Monocular Outdoor Navigation
arXiv:2610.36312v1 Announce Type: new Abstract: Monocular outdoor navigation requires robust camera motion estimation and a dense, navigable map, where 3D Gaussian Splatting SLAM serves this task through an explicit differentiable parameterization and closed-form projection. However, typically, the geometry is recovered by associating 2D pixels with 3D points, so a modular SLAM system absorbs tracking error into the map, leading to cascading degradation. To address these gaps, we present SG^2SLAM, which couples streaming spatial geometry with a ray-conditioned 3DGS kernel. Streaming spatial geometry replaces the per- frame cross-dimensional correspondence with a chunk-level metric prior that supervises tracking and mapping from one source. A ray-conditioned kernel extends each primitive with a viewing-direction subspace that senses the monocular camera viewpoints. The joint optimization over spatial geometry and poses ensures a globally consistent map and trajectory outcome. Experiments on the KITTI and Waymo Open datasets show that our SG^2SLAM surpasses the per-frame association baselines in the monocular outdoor navigation task. Code will be made publicly available upon acceptance.
Generation of quantum light with tunable Wigner negativity with a semiconductor quantum dot
arXiv:2610.36233v1 Announce Type: new Abstract: Wigner-negative states of light are key resources for quantum advantage in continuous-variable optical quantum computing. They can be produced probabilistically via frequency conversion and photon detection, or deterministically by exploiting the single-photon nonlinearity of a quantum emitter. Here, we generate such Wigner-negative states of light with a tunable negativity using a single semiconductor quantum dot in a cavity. Optical coherent control of the quantum-dot transition allows to generate a quantum superposition of zero and one photon. The generated quantum states are characterized point-by-point in the phase space through homodyne measurement. The emergence of Wigner-function negativity is observed as the excitation power is increased. After correcting for collection and propagation losses, we obtain fidelities to the ideal state and quantum purities above 95%, together with near-optimal quadrature squeezing. The ability to generate a wide variety of Wigner-negative states with mature semiconductor technology opens promising opportunities for continuous-variable optical quantum technologies.
Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity
arXiv:2610.33145v1 Announce Type: new Abstract: Understanding how neural activity encodes behavioral and sensory variables is a key challenge in systems neuroscience. This requires models that go beyond predicting activity from labels and instead support source-conditioned counterfactual editing: given a recorded source response, we should be able to edit one measured behavioral or stimulus label and predict how neural activity would change while preserving other observed labels and source-specific variability not explained by the labels. Here, we introduce \textbf{Neural-DISCO}, a disentangled representation learning framework for counterfactual modeling of neural activity. Neural-DISCO partitions the latent space into label-specific components, each associated with an observed behavioral or stimulus label, together with a residual latent that captures variability not supplied through observed labels. Across simulated and real neural recording datasets, we show that this structure enables selective editing of measured labels to generate counterfactual neural responses while preserving the remaining content of the original activity. Further, to gain insight into how behavioral and stimulus labels are encoded at single-neuron resolution, we pair this disentangled framework with feature attribution methods that identify which neurons are most important for each label. We find that disentanglement enhances the recovery of known functional cell roles in both synthetic and real data. Together, these results support Neural-DISCO as a framework for counterfactual modeling of neural activity and provide a practical interface for probing how neural populations encode behavior and sensory information. Code is available at https://github.com/TangLab-UBC/Neural-DISCO-neurips.