AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

LieBN: Batch Normalization over Lie Groups

Ziheng Chen, Yue Song, Rui Wang, Xiao-Jun Wu, Nicu Sebe

Manifold-valued measurements are prevalent in various machine learning tasks. Recent advances have extended Deep Neural Networks (DNNs) to operate on manifolds. These extensions have been accompanied by normalization techniques tailored to different geometries, collectively referred to as Riemannian normalization. However, most existing Riemannian normalization methods are either designed for specific manifolds or fail to effectively normalize manifold-valued sample distributions. To address these limitations, we propose LieBN, a framework for Riemannian Batch Normalization (RBN) over Lie groups. Our approach leverages the theoretically convenient left- and right-invariant metrics, which naturally exist in every Lie group, and provides theoretical guarantees for controlling the Riemannian mean and variance. We instantiate LieBN across nine distinct geometries: four on the Symmetric Positive Definite (SPD) manifold, one on the group of rotation matrices, and four on the manifold of full-rank correlation matrices. Notably, among the SPD metrics, we introduce a novel right-invariant metric and extend three existing Lie group structures via matrix power deformation. Experiments on different manifolds validate the effectiveness of our framework. The code is available at https://github.com/GitZH-Chen/LieBN.git.

arXiv

A Transdiagnostic Space of Disorder Like Phenotypes in Reinforcement Learning Agents

Hari Prasad

Modelling psychological disorders in artificial agents offers a testbed for computational psychiatry and a lens on affective-control failure modes. Prior work induces one or two disorders by hand-tuned reward shaping, labels the behaviour post hoc, and reports single runs. We recast disorder modelling as dose-controllable manipulation of cognitive appraisal signals in an appraisal-guided PPO agent, expressing seven disorders (anxiety, mania, obsessive-compulsive checking, depression, impulsivity, addiction, and post-traumatic stress) each as a single knob grounded in a computational psychiatry account, with each symptom measured by a preregistered assay. Across more than a thousand runs (10 seeds, four controls, 95% confidence intervals) every disorder shows a graded, monotone dose-response that no control reproduces. Beyond these induced effects, three findings emerge that were not written into the reward: disorders self-organise into a two-dimensional affective space in which mania mirrors anxiety; removing a knob remits reward-distortion disorders (mania, checking, addiction) but not avoidance disorders (anxiety, PTSD), which recover under a graded exposure curriculum; and two simultaneous knobs interact nonadditively, yielding testable comorbidity predictions. The depression and addiction knobs further reproduce their double dissociation in a 3D pixel environment (MiniWorld) with a standard convolutional agent and no appraisal critic, showing the framework generalises beyond grid worlds.

arXiv

Context-Masked Truncated Reasoning Audits for Answer-Key Dependence in LLM Tutors

Bonan Shen, Dingyan Shang, Youting Wang, Tao Ning, Bowen Liu

Large language model (LLM) tutors may have access to teacher notes, answer keys, rubrics, or retrieved solutions while producing student-facing explanations. We study whether truncated reasoning probes can distinguish direct access to such private context from answer information carried by the written explanation. Using Truncated Reasoning AUC Evaluation (TRACE), we evaluate 1000 GSM8K problems under question-only, correct answer-key, and wrong answer-key contexts. When forced-answer probes retain the private key, answer-key TRACE AUC rises from 0.375 to 0.900, and the gold answer is recoverable with no explanation at all in 998 of 1000 cases. We then introduce a context-masked replay: answer-key-generated prefixes are probed under the corresponding question-only prompt. Masking reduces 10\% prefix accuracy from 0.997 to 0.126 and median AUC from 0.900 to 0.375, nearly matching question-only values of 0.113 and 0.375. On 746 pairs where both explanations end correctly, the masked mean AUC difference is $-0.0086$ with a 95\% bootstrap interval spanning zero. Wrong keys still account for 272 of 387 incorrect final responses, showing that private artifacts can influence outputs even when early-prefix evidence disappears after masking. These results establish context masking as necessary for attributing early answer availability to an explanation rather than its hidden input.

arXiv

The Theory of Mind Utility: Formal Specification of a Mentalizing Mechanism

Nikolos Gurney, Stacy Marsella

Inferring others' beliefs requires more than reading surface signals; it requires tracking who told them what, in what order, and how credibly. The Theory of Mind Utility (ToM-U) formalizes this epistemic state inference problem at the computational level of analysis, specifying what mentalizing computes and why without commitment to algorithmic or neural implementation. ToM-U achieves this by constructing Local Epistemic World Models (LEWMs) -- directed typed graphs that represent agents, state nodes, and the epistemic relationships among them -- and evaluating discrete candidate LEWMs against observed behavior until one achieves sufficient confidence. Five formal definitions specify the LEWM structure, agent node properties including ordered information access history, a bounded proliferation mechanism for recursive mentalizing, three inference procedures, and a residue function that captures the structured trace left by failed mentalizing attempts. ToM-U differs from Bayesian Theory of Mind and adjacent formal accounts, which presuppose rather than derive belief states, and from simulation theory and theory-theory, which lack a formal apparatus for epistemic state inference. The architecture generates directional, falsifiable predictions about mentalizing failure that follow from structural properties of the model rather than auxiliary assumptions, and positions ToM-U as a domain-agnostic mechanism upstream of goal inference and other downstream social cognitive processes.

arXiv

Reframing AI Loss of Control: What Control Is, How to Have It, How to Lose It

Ze Shen Chin, Maurice Chiodo, Dennis Müller, Coleman Snell

At present, loss of control risks have gained much prominence in public discussion, particularly in relation to AI, with extensive discourse present among academics, frontier labs, and even governments. However, in the existing literature, the concept seems to rest on surprisingly weak foundations, where even those that discuss loss of control extensively do not first establish what control is and what exactly is being lost. Our paper aims to address these gaps. We establish a working definition of control by anchoring it to the "setting and getting of goals". Then, we discuss various aspects of control, built on foundational concepts from related fields like cybernetics, management control, and control theory. This includes who (or what) can be in control, and the things they require to be in control, such as the ability to set goals, having a functional control loop, having requisite variety, and having sufficient goal alignment. Once a framework for control is established, we then discuss how control can be lost, how AIs can contribute to such loss of control, and offer relevant recommendations for how one can maintain control. One interesting consequence of our work is that humanity, as individuals and as groups, can lose varying degrees of control as a result of AI behaviour that is far below the level of superintelligence; the potential for loss of control scenarios (as we define them) already exist, and have existed for a long time.

arXiv

Forensic Trajectory Signatures for Agent Memory Poisoning Detection

Jun Wen Leong

We discover a behavioral invariant in LLM agents under persistent memory poisoning and characterize its deployment boundary. In architectures where retrieval is routed through observable memory-tool invocations, successful attacks require calling memory_recall_fact before email_send_email, a transition mechanistically forced by the attack's information-retrieval dependency. A simple rule exploiting this invariant achieves AUC = 0.9563; a Random Forest over 19 trajectory features refines it to AUC = 0.9904 (BCa 95% CI [0.987, 0.993]). The signature is overdetermined within the poisoned-but-defended evaluation set: removing all recall-related features leaves AUC unchanged. Cross-model hold-out on 9 models (7B-120B) confirms AUC = 1.000 on 6/9 splits, and the invariant transfers to frontier models (GPT-4.1, GPT-4o) without retraining. [v2] A preregistered follow-up (N=4,360, 13 models) reveals a critical deployment boundary: benign memory-grounded sends produce the same recall_before_send signature, yielding 100% false positives conditional on recall_before_send=1 (unconditional benign FPR: 24.7-52.6% depending on recall protocol). The signature is a valid attack precondition, not a maliciousness predicate; standalone blocking is not viable, but gating with recipient metadata restores separation. A prefix-only variant achieves AUC = 0.934, enabling real-time triage.

arXiv

Is Variational Monte Carlo Robust? Sharp Moment Thresholds and Heavy-tailed Stochastic Optimization

Philipp Grohs, Davide Nobile

Variational Monte Carlo (VMC) is a central algorithm in electronic structure theory and has gained renewed importance through modern neural-network ansätze such as FermiNet. At its core, VMC seeks ground states by minimizing the Rayleigh quotient by stochastic optimization. In this work, we show that the resulting stochastic optimization problem is intrinsically governed by the nodal geometry of the underlying wave function. More precisely, we establish that properties of the nodal set determine the integrability of the local energy and gradient estimators that drive VMC. For broad and practically relevant ansatz classes, including Slater-Jastrow wave functions with variable-exponent Slater-type orbitals, we prove that these estimators are generically heavy-tailed and fail to admit higher moments. At the same time, for general analytic ansätze, we prove weak moment bounds for the relevant estimators and identify precise low-moment regimes, showing how generic and degenerate nodal structures lead to different integrability thresholds. Building on this analysis, we introduce a new robust variant of VMC $\unicode{x2013}$ coined PS-Clip-VMC $\unicode{x2013}$ which is based on clipping both the local energies and the per-sample gradients. We prove that PS-Clip-VMC converges both in expectation and with high probability in the weak moment regime of VMC. The robustness of our method is confirmed experimentally by training FermiNet on atoms with up to 18 electrons.

arXiv

Flow-Corrected Thompson Sampling for Non-Stationary Contextual Bandits

Ali Baheri

We study non-stationary linear contextual bandits where the reward model drifts over time, rendering classical contextual bandit algorithms brittle because historical data becomes systematically biased. We propose Flow-Corrected Thompson Sampling (fcTS), a Bayesian method that reuses experience by transporting past rewards to the present using an explicit drift model and incorporating each transported observation with a confidence weight that reflects transport reliability. This yields a unified template that specializes in (i) linear parameter drift via online slope estimation and reward correction, (ii) periodic variation via phase-aware reuse across cycles, and (iii) recurring regime switches via changepoint detection and regime-specific posterior memory. The resulting posterior updates remain closed-form under a linear Gaussian model and can be implemented efficiently with truncated, incrementally updated sufficient statistics. Across five controlled case studies and a semi-synthetic portfolio-selection benchmark with multiple overlapping non-stationarities, fcTS outperforms standard forgetting-based baselines (discounting, sliding windows, and periodic restarts), with the largest gains in settings exhibiting recurring temporal structure. These results demonstrate that when non-stationarity is structured, correcting and reweighting historical observations can be substantially more sample-efficient than uniformly discarding them.

arXiv

The Correctness Illusion in LLM-Generated GPU Kernels

Dipankar Sarkar

Benchmarks for LLM-generated GPU kernels (KernelBench, TritonBench, GEAK) score correctness through fixed-shape, small-sample allclose-style checks. The number of inputs varies between benchmarks. The shape, dtype, and tolerance are fixed for each kernel. We test that oracle empirically. We construct a controlled corpus of 24 Triton and CPU stand-in kernels (15 correct controls and 9 LLM-style buggy variants seeded with documented transcription errors) and re-evaluate it under op-schema-aware seeded fuzzing with a high-precision (fp64) CPU reference and per-(op, dtype) absolute tolerances. The seeded oracle flags 9 of 9 buggy kernels and passes 15 of 15 correct controls, at zero precision cost on controls. We extend the corpus to 26 ops (adding a flash-attention pair) and re-run the same protocol on five GPU classes (RTX 3060, A10, L40S, A100 SXM4, H100 NVL). The verdicts are identical across all five GPUs: 10 of 10 illusions caught and 16 of 16 controls clean. The corpus result is about LLM-style transcription bugs that the allclose-on-one-shape oracle certifies as correct, not about the bug rate of any specific deployed LLM. Every flagged failure replays byte-for-byte from a stored seed.

arXiv

CWind: A Cross-site Router for Large Language Model Inference Serving at Renewable Energy Farms

Tella Rajashekhar Reddy, Atharva Deshmukh, Liangcheng Yu, Chaojie Zhang, Mike Shepperd, Rohan Gandhi, Anjaly Parayil, Srinivasan Iyengar, Ajay Manchepalli, Debopam Bhattacherjee

AI power demand is growing at an unprecedented rate while power grids are often ailing and struggle to keep up. Grid expansion comes with high capital expenditure and long-distance transmission losses, yet there is abundant renewable energy at the source, just not matched to demand. This paper proposes a complementary AI infrastructure deployment model, AI Greeninferencing, that brings modular AI compute to renewable energy sources, focusing on wind, allowing AI footprint expansion, generating local behind-the-meter demand for renewable sites, and helping ease the growing strain on power utilities. Our feasibility analysis shows that 890+ GW of wind capacity lies within 50 ms network round trip time of Azure data centers, and that site-wise right-sizing combined with spatial complementarity of wind energy keeps aggregate fleet utilization on par with traditional deployments. To serve inference requests under variable wind power, we build CWind, a lightweight, reactive, and workload-agnostic AI inference router that uses only real-time signals: inference latency, KV-cache utilization, and queue depth, to dynamically configure sites and distribute requests. Evaluated on a real 64-GPU A100 testbed emulating three wind-powered sites with Azure production traces, CWind reduces P99 end-to-end latency by up to 52% over the strongest contender (also our idea) and by up to 98% over baselines such as power-capping and GPU idling, with consistent gains across workload types, load levels, and GPU generations.

arXiv

The Cognitive Kardashev Scale: Quantifying the Material Envelope of Civilisational Computation

Sachin Sharma

How much thinking can a civilisation do? Kardashev ranked civilisations by the energy they command. This paper borrows his ladder and asks how much machine cognition each rung could support. The arithmetic is deliberately simple. A civilisation has some total power. Only a fraction of that can be spared for computing, and each joule spent buys computation at whatever efficiency the hardware of the day has reached. The product of the three sets a ceiling on machine thought. To keep the resulting quantities intelligible, I express them in units of the human brain's own processing rate, as a rough yardstick rather than a claim about minds. Calibrating the ceiling against present-day supercomputers and AI accelerators led me to two conclusions I did not expect at the outset. Even today's energy supply could support far more machine cognition than humanity actually uses, so physical capacity is not what binds. And whether energy or hardware efficiency becomes the constraint over the coming decade turns on engineering choices that have not yet been made. On the question that may matter most, who gets access to the cognition it describes, the scale is silent. That is a matter of political economy, and the calibration is offered as an input to that debate.

arXiv

Global Automation Atlas

Prashant Garg, Tommaso Crosta, Jasmin Baier

Automation can displace or complement labour, but this need not be constant across economies. Existing exposure measures typically assign fixed scores to tasks or occupations and capture cross-country variation through employment structure. Here we show that feasible automation depends jointly on task content and country-level conditions. We use a large language model to classify 18,797 work tasks in 124 economies by exposure, labour margin, technology channel and artificial-intelligence materiality. Construct-matched components of the measure correlate strongly with established exposure indices, observed work-related ChatGPT use, AI preparedness and firm-reported adoption. The exposed share of tasks ranges from 3.3% to 61.6%, rises with income yet remains heterogeneous within income groups. Lower-income economies are more concentrated in rule-based and labour-substituting forms of automation, whereas physical execution, planning and inference channels, together with labour-augmenting uses of artificial intelligence, become more prominent with development. Country conditioning changes occupation exposure rankings, especially in lower-income economies. Combined with employment data, we find that women are disproportionately employed in occupations with substitution-facing exposure. Machine-learning hypothesis generation identifies digital records, capital equipment, local judgement, trust-based markets and data integration as conditions associated with exposure differences.

arXiv

Sign-Rank, Index, and List Replicability: Connections and Separations

Ari Blondal, Hamed Hatami, Pooya Hatami, Chavdar Lalov, Sivan Tretiak

In learning theory, the sign-rank of a binary concept class captures the smallest dimension in which it can be represented by points and halfspaces. Despite tremendous interest, lower bounds on sign-rank are notoriously difficult to come by. Two recent approaches to the problem establish lower bounds on sign-rank by measures that are easier to analyze: the $\mathbb{Z}_2$-index and the list replicability number. We order these measures, showing that the $\mathbb{Z}_2$-index is upper-bounded by a linear function of the list replicability number. As a main consequence, we obtain a strong separation between sign-rank and $\mathbb{Z}_2$-index, thereby resolving a question of Frick, Hosseini, and Vasileuski. This motivates a thorough study of list replicability, the stronger of the two lower-bounding measures. We establish upper bounds on the list replicability number by two combinatorial measures: height and minimum star number. We also prove a fundamental composition result, showing that the product of two concept classes has list replicability number bounded by the sum of the list replicability numbers of the two classes.

arXiv

AgentJet: A Distributed Swarm Training Framework for Agentic Reinforcement Learning

Qingxu Fu, Boyin Liu, Shuchang Tao, Zhaoyang Liu, Cheng Chen, Xuanfa Jin, Rong Zhu, Bolin Ding

Training reinforcement learning (RL) policies for large language model (LLM) agents requires optimizing multi-turn trajectories that interact with external environments. Existing training frameworks struggle with runtime failures, single-model constraints, incompatible task environments, and redundant context. We present AgentJet, a distributed swarm training framework based on a decoupled multi-node architecture. AgentJet treats the server--client topology as configurable: swarm servers host trainable models and perform optimization on GPU clusters, while detachable swarm clients execute arbitrary agents and communicate through OpenAI-compatible APIs. Reconfiguring this topology supports heterogeneous multi-model RL, mixed-task training with isolated runtimes, fault-tolerant execution, and live code iteration through hot-swappable clients. AgentJet also introduces context tracking with timeline merging, reducing actor-update time by 6.25x on AppWorld. The same detachable-client design supports an automated research system that conducts long-horizon, multi-day RL studies on large-scale clusters with limited human intervention. AgentJet is open-source and compatible with agent systems that issue standard LLM inference requests.

arXiv

Multi$^2$: Hierarchical Multi-Agent Decision-Making with LLM-Based Agents in Interactive Environments

Sangeun Park, Minhae Kwon

A central goal of large language model (LLM) research is to build agentic systems that can plan, act, and adapt through sustained interaction with dynamic environments. While recent LLM-based agents exhibit impressive contextual reasoning, their long-horizon decision-making remains fragile, often suffering from objective drift, where goals and plans drift over extended interactions. We introduce Multi$^2$, a hierarchical multi-agent decision-making framework that explicitly decomposes agent behavior into complementary roles. A high-level agent (System 1) focuses on context-aware sub-goal generation using supervised fine-tuning (SFT), while a low-level agent (System 2) executes atomic actions through offline-to-online reinforcement learning (RL) in interactive environments. This separation enables stable long-horizon control, mitigates objective drift, and allows efficient adaptation. Across diverse interactive environments, Multi$^2$ consistently outperforms strong agentic baselines, demonstrating improved robustness and coordination in multi-turn interaction. Beyond performance, we introduce and release three hierarchical benchmark datasets, filling a long-standing gap in training and evaluating hierarchical decision-making for LLM-based agents.

arXiv

Partial Fusion of Neural Networks: Efficient Tradeoffs Between Ensembles and Weight Aggregation

Fabian Morelli, Stephan Eckstein

Ensembles of neural networks typically outperform individual networks but incur large computational costs, whereas weight aggregation produces less costly, yet also less accurate, aggregate models. We introduce partial fusion of networks, which interpolates between ensembles and weight aggregation and thus allows for a flexible tradeoff between computational cost and performance. A direct way to achieve this is to extend existing weight aggregation methods based on neuron-level similarity between different networks, where partial fusion then only aggregates weights of neurons which are most similar. We showcase one particular method to jointly identify which neurons are most similar and match them via partial optimal transport. Further, we consider the more general perspective of weight aggregation and partial fusion as generalized pruning of ensemble models, where neurons cannot just be deleted, but also linearly combined. Finally, we show that generalized pruning applied to a single network yields similar benefits as partial fusion by allowing for a tradeoff between isolating, deleting, and linearly combining neurons based on similarity. Our code is available at https://github.com/Fabian-Mor/partial_fusion_nn.

arXiv

Bridging the Last Mile of Circuit Design: PostEDA-Bench, a Hierarchical Benchmark for PPA Convergence and DRC Fixing

Pengju Liu, Nuo Xu, Jinwei Tang, Yu Cao, Caiwen Ding

LLM-based agents are increasingly applied to the "last mile" of Electronic Design Automation (EDA): repairing residual sign-off Design Rule Check (DRC) violations and converging Power-Performance-Area (PPA) targets after tool runs. Existing EDA-LLM benchmarks, however, omit DRC fixing entirely and rely on flat hierarchies tied to a single toolchain. We introduce PostEDA-Bench, a hierarchical benchmark with 145 tasks across DRC-Essential, DRC-Reasoning, PPA-Mono, and PPA-Multi, supported by EDA toolchains with machine-checkable evaluation. Across eight commercial and open-source LLMs under multiple agent scaffolds, we find that agents handle synthetic DRC-Essential and single-objective PPA-Mono reasonably well but degrade sharply on the more practical DRC-Reasoning, where the best success rate is 36.66%, and PPA-Multi, where the best success rate is 20.00%; vision augmentation consistently enhances DRC-Bench; and trade-off reasoning, rather than knob knowledge, is the dominant PPA-Multi bottleneck.

arXiv

Agentic AI-assisted coding offers a unique opportunity to instill epistemic grounding during software development

Magnus Palmblad, Jared M. Ragland, Benjamin A. Neely

The capabilities of AI-assisted coding are progressing at breakneck speed. Chat-based vibe coding has evolved into fully fledged AI-assisted, agentic software development using agent scaffolds where the human developer creates a plan that agentic AIs implement. One current trend is utilizing documents beyond this plan document, such as project and method-scoped documents. Here we propose GROUNDING$.$md, a community-governed, field-scoped epistemic grounding document, using mass spectrometry-based proteomics as an example. We have drafted a file specific to proteomics: proteomics_GROUNDING$.$md (available at https://github.com/OmicsGrounding/proteomics-grounding) to demonstrate what a GROUNDING$.$md would look like. This explicit field-scoped grounding document encodes Hard Constraints (non-negotiable validity invariants empirically required for scientific correctness) and Convention Parameters (community-agreed defaults) that override all other contexts to enforce validity, regardless of what the user prompts. In practice, this will empower a non-domain expert to generate code, tools, and software that have best practices baked in at the ground level, providing confidence to the software developer but also to those reviewing or using the final product. Undoubtedly it is easier to have agentic AIs adhere to guidelines than humans, and this opportunity allows for organizations to develop epistemic grounding documents in such a way as to keep domain experts in the loop in a future of democratized generation of bespoke software solutions.

arXiv

GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding

Fanxu Meng

Multi-head Latent Attention (MLA), the attention used in DeepSeek-V2/V3, jointly compresses keys and values into a low-rank latent and matches the H100 roofline almost perfectly. Its trained weights, however, expose only one decoding path - an absorbed MQA form - which ties efficient inference to H100-class compute-bandwidth ratios, forfeits tensor parallelism along the head axis, and yields no Multi-Token Prediction (MTP) gain on commodity inference GPUs such as the export-restricted H20. We propose Group-Query Latent Attention (GQLA), a minimal modification of MLA whose trained weights expose two algebraically equivalent decoding paths over the same parameters: an MQA-absorb path identical to MLA's, and a GQA path with a per-group expanded cache. The runtime picks the path that matches the target hardware - no retraining, no custom kernels - so a single set of GQLA weights pins the rooflines of both H100 (MQA-absorb, s_q=1) and H20 (GQA + MTP, s_q=2), while supporting up to 8-way zero-redundancy tensor parallelism on the GQA path. To avoid pretraining from scratch we extend TransMLA into TransGQLA, which converts a pretrained GQA checkpoint into a GQLA model; on LLaMA-3-8B it compresses the per-token KV cache to 28.125% of the GQA baseline on the MQA-absorb path while structurally preserving GQA-level traffic on the per-group path.

arXiv

How Many Iterations to Jailbreak? Dynamic Budget Allocation for Multi-Turn LLM Evaluation

Shai Feldman, Yaniv Romano

Evaluating and predicting the performance of large language models (LLMs) in multi-turn conversational settings is critical yet computationally expensive; key events -- e.g., jailbreaks or successful task completion by an agent -- often emerge only after repeated interactions. These events might be rare, and under any feasible computational budget, remain unobserved. Recent conformal survival frameworks construct reliable lower predictive bounds (LPBs) on the number of iterations to trigger the event of interest, but rely on static budget allocation that is inefficient in multi-turn setups. To address this, we introduce \emph{Dynamic Allocation via PRojected Optimization} (DAPRO), the first theoretically valid dynamic budget allocation framework for bounding the time-to-event in multi-turn LLM interactions. We prove that DAPRO satisfies the budget constraint and provides distribution-free, finite-sample coverage guarantees without requiring the conditional independence between censoring and event times assumed by prior conformal survival approaches. A key theoretical contribution is a novel coverage bound that scales with the square root of the mean censoring weight rather than the worst-case weight, yielding provably tighter guarantees than prior work. Furthermore, DAPRO can be employed to obtain unbiased, low-variance estimates of population-level evaluation metrics, such as the jailbreak rate, under limited computing resources. Comprehensive experiments across agentic task success, adversarial jailbreaks, toxic content generation, and RAG hallucinations using LLMs such as Llama 3.1 and Qwen 2.5 demonstrate that DAPRO consistently achieves coverage closer to the nominal level with lower variance than static baselines, while satisfying the budget constraint.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.