Conservatism

Policy learning aims to determine who should be treated based on individual characteristics. In high-stakes settings such as medicine and public policy where safety is a central concern, improving the average outcomes alone may not be sufficient: decision makers may also seek to protect individuals from harm, in line with the Hippocratic principle of ``do no harm.'' In this paper, we propose \textit{conformal policy learning} (CPL), a policy learning procedure with a new distribution-free safety guarantee that controls the probability of assigning treatment to an individual who would be harmed relative to control. CPL views each treatment decision as testing a hypothesis of counterfactual harm and assigns treatment by thresholding conformal p-values. These p-values use observable proxies and selective calibration to address the challenge that the potential outcomes under comparison are never simultaneously observed. For randomized experiments, under standard exchangeability conditions, CPL provides finite-sample safety guarantee at a user-specified level, without imposing any outcome modeling assumptions. Moreover, when the outcome model is consistently estimated, CPL achieves asymptotically optimal welfare subject to the safety constraint. In observational studies, CPL with learn-then-balance weights achieves doubly robust safety guarantees. We evaluate CPL through extensive simulations and apply it to an empirical study of AI-powered interventions designed to reduce conspiracy beliefs.

Harvard Dataverse hosts over 150,000 research datasets, but the geographic information those datasets carry is entered as free text by depositors and has never been assembled into a searchable structure. We construct a knowledge graph from the repository's public data and metadata, organizing 102,650 datasets within a 215,985-node network of 528,003 edges linking datasets to keywords, publications, subjects, journals, and locations. Of those datasets, 43,991 (42.9 percent) carry at least one geospatial field, geographic coverage, geographic unit, or a bounding box and 96.9 percent of all nodes sit in a single connected component, so datasets remain reachable from one another even when their geospatial metadata share nothing in common. A conservative keyword search identifies 7,654 geospatially tagged datasets (17.4 percent) as directly policy-relevant, with elections and legislatures the largest cluster, followed by government administration, health policy, transportation, and education. Five datasets illustrate how this metadata behaves across policy domains and spatial scales, and an extended use case shows how community language models, stance detection with geographic aggregation, and partisan language bridging tools can attach discourse to place. The central obstacle is place resolution: the same location appears as many disconnected nodes. We argue that the graph provides a concrete setting for developing AI-driven metadata enrichment and entity resolution, and we document its coverage skew toward American, city-level data.

Physical learning lets a trainable material or network use its own physical response to carry error signals, reducing the need for a separately programmed backward computation. We ask what determines whether such a system follows conventional gradient descent or evolves along a genuinely different learning trajectory. Our canonical model is a directed layered transport network in which every node redistributes a fixed amount of flow, so learning preserves positivity and total mass. In this model, conservation constrains only the allowable learning directions. Within the matched response class studied here, adjoint matching gives the physical output response a symmetric form. Non-negative mode-wise feedback then produces a reciprocal closed-loop response and a reweighted gradient flow. Adding an antisymmetric boundary component makes the closed-loop response rotational: the learning path can turn while the error driving that update still decreases at that moment. Turning is not automatically beneficial. Its finite-step effect is set by local curvature, and its accumulated effect also depends on step selection and on the new states visited along the path. Numerical consistency checks reproduce the exact response structure, predict the sign of the local effect across new network families, and show how trajectory drift can negate a local advantage. These results separate the roles of conservation, reciprocity, and nonreciprocity in physical learning.

Physical learning rules such as equilibrium propagation (EP), coupled learning (CL), and adjoint coupled learning (AL) train resistive networks through local measurements. In the small-nudge limit EP and CL exactly conserve the conductance mass K = (1/2) sum_e kappa_e^2, a property that stabilizes training. We show that conservation also governs the inductive bias of these rules. For a single output we prove that EP and CL are trajectory equivalent, so single-output experiments cannot distinguish what the two rules learn. We prove that AL does not conserve the mass but dissipates it at exactly twice its own loss. In linear circuits we prove that the conserved mass has no functional consequence: all three vector fields are homogeneous in the conductances, so the selected solution is independent of the initialization scale. Fixed nonlinear elements break this protection. In diode circuits the learned input-output function depends on the initialization scale by up to about forty percent, an effect absent in linear controls, and the conservative rules retain this memory permanently while the dissipative rule partially erases it. At matched training loss the dissipative rule typically generalizes worse than the conservative rules, although it reaches low training loss faster; the penalty correlates with the mass dissipated en route and fades in larger circuits, where little mass is lost. The conservation structure of a local learning rule thus sets its initialization memory, its training speed, and, where dissipation is appreciable, its generalization; it should be treated as a design parameter of physical learning machines.

Jailbreak defenses are essential for protecting large language models (LLMs), but they can also introduce secondary costs that weaken model utility. We present a systematic study of these defense trade-offs along three dimensions: performance impact, over-refusal on benign inputs, and inference cost. Rather than treating defenses as a single class, we organize them by operational strategy and examine how different strategies correlate with different side-effect profiles. Across state-of-the-art defense methods, widely used benchmark datasets, and representative open-source LLMs, we find that defenses rarely improve downstream capability, but instead vary in how they trade safety gains against usability and efficiency. In particular, rule-based defenses best preserve task performance, highly conservative self-reflective defenses often increase over-refusal, and multi-round defenses incur the largest runtime overhead. These results provide both a benchmark for evaluating defense side effects and practical guidance for selecting defenses under deployment constraints.

Offline reinforcement learning (RL) aims to learn an effective policy from a static dataset, but its performance is fundamentally limited by dataset coverage. Action preference queries leverage expert feedback without additional environment interaction, enabling policy improvement during offline training. However, existing methods still face two key challenges: selecting informative preference queries and effectively exploiting the collected feedback. Current approaches typically rely only on the distance between policy actions and dataset actions for query selection, while enforcing fixed constraints that keep the policy close to queried preferences. Such strategies often lead to unstable policy updates and integrate poorly with value regularization. To address these limitations, we propose Conservative Query and Adaptive Regularization under Uncertainty Estimation, a lightweight framework that jointly improves preference querying and preference exploitation. Specifically, we employ a Morse network to estimate the uncertainty of policy actions with respect to the offline dataset. Based on this uncertainty, we introduce a conservative query strategy that selectively queries actions near the dataset to preserve Bellman-update stability, together with an uncertainty-aware adaptive regularization scheme that dynamically adjusts data-level constraints during policy optimization. We integrate our framework with CQL and evaluate it extensively on the D4RL benchmark. Experimental results demonstrate superior or competitive performance across a wide range of tasks.

Cross-subject generalization remains a fundamental challenge in surface electromyography (sEMG)-based gesture recognition. Although deep learning methods have improved within-subject performance, they often rely on subject-specific data and struggle to balance invariance and discriminability. In this work, we propose a conservative multi-objective learning framework for subject-invariant sEMG gesture recognition. The proposed model adopts a multi-head architecture that jointly optimizes gesture classification, adversarial subject confusion through gradient reversal, and triplet-based metric learning to encourage discriminative and subject-invariant representations. To improve optimization stability, a Lipschitz-inspired adaptive weighting mechanism is introduced to dynamically balance the auxiliary objectives according to their relative magnitudes during training. The proposed method is evaluated on two benchmark datasets: UCI EMG (36 subjects, 6 gestures) and NinaPro DB5 (10 subjects, 10 gestures). On the UCI EMG dataset, the method achieves 84.48% accuracy compared to 78.2% reported by state-of-the-art methods. On NinaPro DB5, it achieves 61.44% accuracy versus 41.30%, corresponding to a 49% relative improvement. In addition, the proposed framework reduces cross-subject prediction variance and produces more structured latent representations. These results indicate that jointly enforcing invariance and discriminability through adaptive multi-objective optimization leads to more stable training and improved cross-subject generalization in sEMG-based gesture recognition systems.

Conservative offline training is widely advocated as a safe foundation for subsequent online adaptation: if a policy stays close to well-supported behaviour, the argument goes, it is less likely to exploit imperfections in a learned reward model. We challenge this intuition empirically and mechanistically. We train a Qwen3-14B policy under Direct Preference Optimisation (DPO) with three levels of conservatism ($β\in \{β_{\mathrm{lo}}, β_{\mathrm{mid}}, β_{\mathrm{hi}}\}$ derived from empirical log-ratio percentiles), then adapt each checkpoint online against a learned reward ensemble (3,$\times$,Qwen3-1.7B) while measuring true performance on GSM8K exact-answer accuracy. We find that \emph{higher offline conservatism monotonically increases reward-hacking damage}, measured by the Goodhart gap and its area under the curve (AUGC), with Spearman $ρ= 1.0$ across all three conditions. Mechanistic analysis reveals a three-link causal chain: (i) high-$β$ DPO compresses policy entropy, (ii) Low-entropy policies generate responses with reduced diversity, concentrating in a narrow region of the reward model's training distribution (lower pairwise cosine distance), and (iii) despite this proximity, ensemble disagreement (epistemic uncertainty) increases with $β$ and is exploited faster during online optimisation. We further fit a power-law curve to the $(β, \augc)$ data and identify a practical optimal conservatism level $β^{\star}$ that balances alignment fidelity against hacking vulnerability. Our results suggest that the field needs \emph{calibrated}, not \emph{maximal}, conservatism.

We ask a representation-learning question about physical world models: when does a conservation law remain certifiable after a model learns a latent representation? A certified horizon bounds -- in advance, from measurable model defects -- how many steps a rollout provably stays on a physical invariant's level set. The key design choice is what is certified: not a learned latent Hamiltonian or a learned scalar witness (a model can conserve either while drifting in true energy), but the decoded physical invariant obtained by decoding the latent state and evaluating the known invariant. Around this object we derive shell-horizon certificates whose budget decomposes into representation, readout, and latent-dynamics defects, with a monotone alignment bridge through which a soft learned witness yields a certified horizon for the decoded invariant, and test them across state, learned-lift, and pixel observations on conservative systems. Conservation certificates can survive learned representation, but not all geometric priors survive equally. Hard canonical symplectic structure yields the longest horizons in known phase coordinates yet does not cross a learned chart, whereas a controlled-Lipschitz-aligned soft invariant survives in the nonlinear learned-representation settings we test -- two lift systems, with the gain growing with nonlinearity, and pixels. Pixel certification is recovered on a readout-stable sub-tube, and the Kepler problem exposes a geometric boundary. The central object is therefore not a latent Hamiltonian, but a decoded physical invariant whose robustness to representation learning can be measured, certified, and falsified.

We identify and formally characterize credit misassignment as a systematic failure mode of GRPO in tool-augmented multimodal search agents: its uniform broadcast of trajectory-level advantages to all tokens causes valuable tool-use steps in failing trajectories to be penalized no differently from valueless ones. We further empirically quantify the scale of this phenomenon. Over half of failing trajectories and failing tool-use actions exhibit correctable credit misassignment, demonstrating that the wasted training signal is both substantial and structurally exploitable. Building on this insight, we propose Tool-Aware Policy Optimization (TAPO), which exploits the parameter-determinism property of information-acquisition tools: similar call parameters define equivalent information-acquisition actions and should therefore share comparable action credit. TAPO constructs counterfactual witnesses within the current training batch and compensates misassigned negative credit via confidence-gated conservative advantage correction. It requires no additional annotation, models, or sampling, and introduces negligible computational overhead. Across multiple multimodal search benchmarks, TAPO delivers consistent, plug-and-play improvements over strong baselines for three mainstream RL algorithms (GRPO, GSPO, and SAPO). Our code and models will be publicly released upon acceptance.

Financial decision-making tasks such as stock recommendation and portfolio allocation typically estimate future return and risk and then select trades or allocations for an investor, and the chosen optimization objective often determines realized performance. However, because market conditions evolve over time, a fixed objective can be suboptimal across regimes, while regime-switching pipelines that rely on latent regime estimates can be noisy or delayed and frequent switching can increase turnover and operational instability. In this paper, we propose DOSS (Dynamic Objective Selection with Safeguards), a learning-based selector that directly chooses the decision-relevant objective function at each time point from interpretable statistical summaries of recent returns, selecting among a small set of candidates (e.g., return-seeking, loss-averse, and risk-adjusted) without introducing intermediate regime variables. DOSS formulates objective selection as a classification problem over objectives and performs sequential updates with a rolling window to make forward-looking selections without temporal leakage, while also outputting a confidence score for each proposal. To mitigate misselection and excessive switching in deployment, DOSS applies confidence-aware gating with a fail-safe that overrides low-confidence proposals to a conservative default and enforces explicit controls tied to switching frequency. We further integrate governance by positioning a Large Language Model (LLM) as an oversight component rather than a generator of new objectives: the LLM is restricted to accept a proposed objective or override it to a predefined safe default, with deterministic rule-based constraints triggering overrides when needed.

We present an online, distribution-free framework for controlling the Conditional Value-at-Risk (CVaR), extending conformal tail risk control to non-stationary and adversarial environments. Unlike classical risk control methods, which rely on stationarity or linearity of expectation, our approach provides provable safety guarantees for a nonlinear tail risk functional under arbitrary data-generating processes that may drift or shift strategically over time. By leveraging deep connections between conformal tail risk control, online learning, and the variational representation of CVaR introduced by Rockafellar and Uryasev, we develop a novel procedure for online CVaR control with adversarial regret guarantees. The proposed method operates without assumptions on the underlying data-generating process, making it broadly applicable in modern high-stakes deployment settings. We prove that the realized empirical CVaR is asymptotically controlled at the target level, and that the resulting control is asymptotically tight up to a finite-sample conservatism gap. We demonstrate the effectiveness of our approach on portfolio risk management and toxicity mitigation for Large Language Models (LLMs), where rare but catastrophic failures dominate system risk.

Offline reinforcement learning requires careful conservatism to mitigate distribution shift, yet most existing methods apply a fixed penalty uniformly across all states regardless of local data coverage. We present UNIQ (Uncertainty-Informed Quantile), an offline RL method that introduces state-adaptive conservatism through conformally calibrated uncertainty estimation. Built on the Implicit Q-Learning (IQL) backbone, UNIQ trains a multi-expectile value ensemble, computes distribution-free uncertainty estimates using split conformal prediction, and maps the resulting signal to a state-dependent expectile that relaxes conservatism in well-covered regions while strengthening it in uncertain regions near the data frontier. On D4RL MuJoCo benchmarks, UNIQ consistently improves over IQL, with the largest gains observed on Walker2d and replay-heavy tasks. At the same time, UNIQ operates at near-IQL memory cost (approximately 250 MB peak VRAM), providing roughly a 10x reduction compared to EDAC. Rather than pursuing overall state-of-the-art performance, we position UNIQ as a practical mechanism contribution that improves the performance-efficiency trade-off in offline reinforcement learning.

Agentic AI systems capable of autonomous planning and extended environmental interaction pose a fundamental control problem: how can humans maintain meaningful oversight of systems that may exceed their own capabilities? Existing approaches to scalable oversight rely on complex assumptions, remain largely heuristic, or lack practical methods for sequential settings with statistical guarantees. We introduce Calibrated Collective Oversight (CCO), which aggregates diverse auxiliary scoring functions into a penalty measuring deviation from a conservative baseline. Inspired by Attainable Utility Preservation, CCO enables collective conservatism: actions face a penalty proportional to overseer concern, so high-utility actions are still selected when overseers find them unobjectionable and overridden only when concern accumulates. CCO calibrates this conservatism online using Conformal Decision Theory, ensuring that undesirable outcomes remain below a user-specified target threshold with finite-time bounds and no distributional assumptions. On a modified version of SWE-bench, weaker overseers successfully constrain an adversarially misaligned stronger agent; on MACHIAVELLI, CCO substantially reduces ethical violations while preserving reward. In both settings, empirical violation rates closely match the specified targets, as predicted by the theory.

We propose and analyze a conservative drifting method for one-step generative modeling. The method replaces the original displacement-based drifting velocity by a kernel density estimator (KDE)-gradient velocity, namely the difference of the kernel-smoothed data score and the kernel-smoothed model score. This velocity is a gradient field, addressing the non-conservatism issue identified for general displacement-based drifting fields. We prove continuous-time finite-particle convergence bounds for the conservative method on $\R^d$: a joint-entropy identity yields bounds for the empirical Stein drift, the smoothed Fisher discrepancy of the KDE, and the squared center velocity. The main finite-particle correction is a reciprocal-KDE self-interaction term, and we give deterministic and high-probability local-occupancy conditions under which this term is controlled. We keep the quadrature constants explicit and track their possible bandwidth dependence: the root residual-velocity rate $N^{-1/(d+4)}$ holds under an additional $h$-uniform quadrature regularity condition, while a more general growth condition yields the optimized root rate $N^{-(2-β)/(2(d+4-β))}$, where $0\le β<2$. We also analyze the non-conservative drifting method with Laplace kernel, corresponding to the original displacement-based velocity proposed in Deng et al., 2026 (arxiv:2602.04770). For this method, a sharp companion kernel decomposes the velocity into a positive scalar preconditioning of a sharp-score mismatch plus a Laplace scale-mismatch residual, producing an analogous finite-particle rate with an unavoidable residual term. Finally, we explain how the continuous-time residual-velocity bounds translate into one-step generation guarantees through the explicit drift size $η$.

Logged advertising auctions make offline reserve-price evaluation attractive but risky. Replay tables can identify policies with large apparent yield gains, yet they can also hide weak threshold support, multiple-comparison effects, subgroup harm, and bidder-response uncertainty. Existing replay and off-policy evaluation methods estimate or rank policy values, but they do not directly answer the operational question of whether the available evidence is strong enough to justify validation. This paper develops a support-aware offline decision framework for reserve-policy selection. Rather than outputting a single point-estimate winner, the framework converts logged evidence into a conservative decision object consisting of certified policies, statistically dominated alternatives, and unresolved candidates requiring further validation. The main theoretical result gives a unified finite-catalog guarantee showing that, under simultaneous uncertainty control and conservative support gates, the framework preserves the best gate-passing policy while eliminating only policies with certified regret. Supporting results characterize support-localized replay generalization, establish information-theoretic threshold-resolution limits, and quantify when heterogeneous bidder response can overturn localized replay rankings. Experiments on iPinYou real-time-bidding logs show that the leading reserve rule achieves a 47.66% replay lift in season two, a 40.71% simultaneous lower-bound lift, and a 43.87% frozen out-of-time replay lift in season three. The framework reduces a 19-policy catalog to a two-policy validation shortlist while certifying non-harm across 44 advertiser, exchange, and region segments. The results support the central claim that offline reserve-policy evaluation should produce certified validation decisions rather than point-estimate rankings alone.

While Large Language Models (LLMs) are increasingly integrated into in-vehicle conversational systems, identifying the optimal model remains challenging due to the lack of domain-specific evaluation standards tailored to real-world deployment requirements. In this paper, we propose a novel evaluation framework for in-vehicle assistants, with a particular focus on Korean-language localization. Our empirical analysis reveals notable patterns in model behavior. First, fine-grained Korean honorific control remains unstable in current LLMs, indicating that precise speech-level realization must be explicitly evaluated in localization settings. Second, models exhibit weaker performance in strategic conversational metrics like clarification and proactivity. Our analysis suggests this stems from the inherent subjective complexity of these tasks, where our framework adopts a conservative evaluation stance to prioritize reliability. Together, our findings underscore that automotive AI must move beyond general competence toward precise linguistic tailoring and reliable, safety-oriented interaction management.

Media bias detection has predominantly been framed as a classification task: assign a political label to an article or outlet. We argue this framing is too shallow: it identifies that bias exists but not where, how, or crucially, what is structurally omitted. We present NewsLens, a five-agent adversarial pipeline for structured news bias navigation. A Fact Verifier, Progressive Framing Analyst, Conservative Framing Analyst, Propaganda Detector, and Neutral Summarizer collaborate to deconstruct articles into interpretable framing maps, exposing ideological omissions, rhetorical manipulation, and framing boundaries. The system is evaluated on 15 articles across four geopolitical event clusters (India-Pakistan Kashmir, Gaza, Climate Policy, Ukraine) using Qwen2.5-3B-Instruct (4-bit quantised, Google Colab T4), with cross-model validation using Mistral 7B on the Kashmir cluster. Center outlets show the highest mean Perspective Divergence Score (PDS: Qwen 0.907, Mistral 0.729 on Kashmir subset); conservative-framing outlets show the highest mean Manipulation Index (MI: 0.600 across both models). Cross-model comparison shows high consistency for high-propaganda content (Republic World delta-PDS=0.125, MI=0.8 both models) and greater variance for nuanced reporting. Mann-Whitney U tests find no statistically significant between-group differences at n=15, reported honestly as a sample-size limitation confirmed by post-hoc power analysis. A partial ablation removing the Propaganda Detector shows degraded omission precision in the Neutral Summarizer output. The architecture extends prior lexical-geometric bias work to agentic LLM reasoning, and is fully reproducible using open-weight models without API keys.

Uncertainty quantification is essential in safety-critical settings--from autonomous driving to aviation, finance, and health--where decisions must rely on conservative bounds rather than point estimates. Predictor-level intervals (e.g., from quantile regression, conformal prediction, variance networks, or Bayesian models) generally do not compose: adding two per-variable intervals need not yield a valid interval for their sum or preserve coverage. In aviation, Gaussian overbounding replaces complex error distributions with a conservative Gaussian whose tails dominate the truth, so conservatism propagates through linear operations. Yet classical overbounds are global, often overly conservative, and hard to adapt to feature-conditioned errors. We propose a unified learning framework that trains neural networks to produce context-aware Gaussian overbounds--mean and scale--with provable conservatism on a finite quantile grid and, under three explicit regularity assumptions, continuous-tail conservatism on a certified interval. Our overbounding loss enforces conservativeness at selected quantiles while penalizing distributional distance with a Wasserstein-style term. The learned bounds support conservative linear-combination and convolution analysis on the enforced grid, and on the certified interval when assumptions hold, while being less redundant than traditional methods. We provide a scoped analysis of discrete-to-continuous conservatism and compact-domain objective regularity, and validate on synthetic data and real-world datasets, including multipath, ionospheric, and tropospheric residual errors. Across these settings, the method yields tighter bounds while maintaining conservatism on the enforced grid and in experiments. The framework is modality-agnostic and applicable to learning systems that require conservative, feature-conditioned uncertainty estimates in dynamic environments.

As AI agents move from chat interfaces to systems that read private data, call tools, and execute multi-step workflows, guardrails become a last line of defense against concrete deployment harms. In these settings, guardrail failures are no longer merely answer-quality errors: they can leak secrets, authorize unsafe actions, or block legitimate work. The hardest failures are often contextual: whether an action is acceptable depends on local privacy norms, organizational policies, and user expectations that resist pre-deployment specification. This creates a practical gap: guardrails must adapt to their own operating environments, yet deployment feedback is typically limited to sparse, noisy user-reported failures, and repeated fine-tuning is often impractical. To address this gap, we propose LiSA (Lifelong Safety Adaptation), a conservative policy induction framework that improves a fixed base guardrail through structured memory. LiSA converts occasional failures into reusable policy abstractions so that sparse reports can generalize beyond individual cases, adds conflict-aware local rules to prevent overgeneralization in mixed-label contexts, and applies evidence-aware confidence gating via a posterior lower bound, so that memory reuse scales with accumulated evidence rather than empirical accuracy alone. Across PrivacyLens+, ConFaide+, and AgentHarm, LiSA consistently outperforms strong memory-based baselines under sparse feedback, remains robust under noisy user feedback even at 20% label-flip rates, and pushes the latency--performance frontier beyond backbone model scaling. Ultimately, LiSA offers a practical path to secure AI agents against the unpredictable long tail of real-world edge risks.

Optimal control for safety-critical systems is often dependent on the conservativeness of constraints. Control Barrier Functions (CBFs) serve as a medium to represent such constraints, but constructing a minimally conservative CBF is a computationally intractable problem. Therefore, approaches that can guarantee safety while reducing conservatism will help improve the optimality of the system under consideration. Here, we present a Model Predictive Control (MPC) formulation using CBF as a terminal constraint, which is proven to improve feasibility and reachable sets with increasing prediction horizon. The constructive nature of the proofs allows for warm-starting the nonlinear optimization problem, thereby reducing the computational time substantially. Simulations are set up for a simple nonholonomic system to numerically validate the results, and it is observed that the number of infeasible points decreased by a factor of 1.7 to 2.7. The increase in reachable state space was demonstrated by the ability of the system to track trajectories that are entirely inside the unsafe region of the control barrier function.

We revisit a classic "chicken-and-egg" problem in reinforcement learning: to safely improve a policy, the value function must be accurate on the state-visitation distribution of the updated policy. That distribution over states is unknown and cannot be sampled for the purposes of training the value function. Conservative updates solve this problem, but at the cost of shrinking the policy update. This paper explores an alternative solution, Approximate Next Policy Sampling (ANPS), which addresses the problem by modifying the training distribution rather than constraining the policy update. ANPS is satisfied if the distribution of the training data approximates that of the next policy. To demonstrate the feasibility and efficacy of ANPS, we introduce Stable Value Approximate Policy Iteration (SV-API). SV-API modifies the standard approximate policy iteration loop to hold the target policy fixed while an iteratively updated behavioral policy gathers relevant experience. It only commits to a new policy once a convergence criterion has been met. If certain stability criteria are met, the update is guaranteed to be safe; otherwise, it remains no less safe than standard approximate policy iteration. Applying SV-API to PPO yields Stable Value PPO (SV-PPO), which matches or improves performance on high-dimensional discrete (Atari) and continuous control benchmarks while executing substantially larger target policy updates. These results demonstrate the viability of ANPS as a new solution to this classic challenge in RL.

Variable impedance model predictive control (MPC) formulations often treat joint stiffness as an instantaneous decision variable. The resulting feasible set strictly contains the physically realizable set under first-order actuator dynamics. We identify this as a formulation error rather than a modeling approximation, formalize the distinction between the parameter-based feasible set F_param and the realizable set F_real, and characterize the regime of mismatch via the dimensionless parameter α = ωsT (actuator bandwidth times task timescale). For the 1D hopping monoped, we prove that below an analytical threshold α_crit derived in closed form from task physics, no admissible stiffness command realizes the parameter-based prediction. Numerical validation in 1D shows monotonic deviation growth as α decreases, with the predicted scaling holding across ten parameter combinations (log-log R2 = 0.986). Mechanism transfer to planar spring-loaded inverted pendulum dynamics confirms center-of-mass and stance-timing deviation as the primary consequence, with regime-dependent friction effects as a tertiary observable. A second threshold α_infeas < α_crit establishes a floor below which restricting the admissible stiffness range cannot repair realizability, closing the conservative-tuning objection. Augmenting the prediction state with stiffness closes the mismatch by construction.

Vision-language models (VLMs) have demonstrated remarkable capabilities in understanding human-object interactions, but their application to robotic systems with non-humanoid morphologies remains largely unexplored. This work investigates whether VLMs can effectively infer affordances for robots with fundamentally different embodiments than humans, addressing a critical gap in the deployment of these models for diverse robotic applications. We introduce a novel hybrid dataset that combines annotated real-world robotic affordance-object relations with VLM-generated synthetic scenarios, and perform an empirical analysis of VLM performance across multiple object categories and robot morphologies, revealing significant variations in affordance inference capabilities. Our experiments demonstrate that while VLMs show promising generalisation to non-humanoid robot forms, their performance is notably inconsistent across different object domains. Critically, we identify a consistent pattern of low false positive rates but high false negative rates across all morphologies and object categories, indicating that VLMs tend toward conservative affordance predictions. Our analysis reveals that this pattern is particularly pronounced for novel tool use scenarios and unconventional object manipulations, suggesting that effective integration of VLMs in robotic systems requires complementary approaches to mitigate over-conservative behaviour while preserving the inherent safety benefits of low false positive rates.

Fine-grained medical image classification is challenged by subtle inter-class variations and visually ambiguous cases, where confidence estimates often exhibit uncertainty rather than being overconfident. In such scenarios, purely discriminative classifiers may achieve high overall accuracy yet still fail to distinguish between highly similar categories, leading to miscalibrated predictions. We propose T-DuMpRa, a teacher-guided dual-path multi-prototype retrieval-augmented framework, where discriminative classification and multi-prototype retrieval jointly drive both training and prediction. During training, we jointly optimize cross-entropy and supervised contrastive objectives to learn a cosine-compatible embedding geometry for reliable prototype matching. We further employ an exponential moving average (EMA) teacher to obtain smoother representations and build a multi-prototype memory bank by clustering teacher embeddings in the teacher embedding space. Our framework is plug-and-play: it can be easily integrated into existing classification models by constructing a compact prototype bank, thereby improving performance on visually ambiguous cases. At inference, we combine the classifier's predicted distribution with a similarity-based distribution computed via cosine matching to prototypes, and apply a conservative confidence-gated fusion that activates retrieval only when the classifier's prediction is uncertain and the retrieval evidence is decisive and conflicting, otherwise keeping confident predictions unchanged. On HAM10000 and ISIC2019, our method yields 0.68%-0.21% and 0.44%-2.69% improvements on 5 different backbones. And visualization analysis proves our model can enhance the model's ability to handle visually ambiguous cases.

An agent must try new behaviors to explore and improve. In high-stakes environments, an agent that violates safety constraints may cause harm and must be taken offline, curtailing any future interaction. Imitating old behavior is safe, but excessive conservatism discourages exploration. How much behavior change is too much? We show how to use any safe reference policy as a probabilistic regulator for any optimized but untested policy. Conformal calibration on data from the safe policy determines how aggressively the new policy can act, while provably enforcing the user's declared risk tolerance. Unlike conservative optimization methods, we do not assume the user has identified the correct model class nor tuned any hyperparameters. Unlike previous conformal methods, our theory provides finite-sample guarantees even for non-monotonic bounded loss functions, and it introduces a new policy control setting. Our experiments on applications ranging from natural language question answering to biomolecular engineering show that safe exploration is not only possible from the first moment of deployment, but can also improve performance.