In many practical sequential decision-making problems, tracking the state of the environment incurs a sensing/communication/computation cost. In these settings, the agent's interaction with its environment includes the additional component of deciding when to sense the state, in a manner that balances the value associated with optimal (state-specific) actions and the cost of sensing. We formulate this as an expected discounted cost Markov Decision Process (MDP), wherein the agent incurs an additional cost for sensing its next state, but has the option to take actions while remaining `blind' to the system state. We pose this problem as a classical discounted cost MDP with an expanded (countably infinite) state space. While computing the optimal policy for this MDP is intractable in general, we derive lower bounds on the optimal value function, which allow us to bound the suboptimality gap of any policy. We also propose a computationally efficient algorithm SPI, based on policy improvement, which in practice performs close to the optimal policy. Finally, we benchmark against the state-of-the-art via a numerical case study.

Real-world language agents must handle complex, multi-step workflows across diverse Apps. For instance, an agent may manage emails by coordinating with calendars and file systems, or monitor a production database to detect anomalies and generate reports following an operating manual. However, existing language agent benchmarks often focus on narrow domains or simplified tasks that lack the diversity, realism, and long-horizon complexity required to evaluate agents' real-world performance. To address this gap, we introduce the Tool Decathlon (dubbed as Toolathlon), a benchmark for language agents offering diverse Apps and tools, realistic environment setup, and reliable execution-based evaluation. Toolathlon spans 32 software applications and 604 tools, ranging from everyday platforms such as Google Calendar and Notion to professional ones like WooCommerce, Kubernetes, and BigQuery. Most of the tools are based on a high-quality set of Model Context Protocol (MCP) servers that we may have revised or implemented ourselves. Unlike prior works, which primarily ensure functional realism but offer limited environment state diversity, we provide realistic initial environment states from real software, such as Canvas courses with dozens of students or real financial spreadsheets. This benchmark includes 108 manually sourced or crafted tasks in total, requiring interacting with multiple Apps over around 20 turns on average to complete. Each task is strictly verifiable through dedicated evaluation scripts. Comprehensive evaluation of SOTA models highlights their significant shortcomings: the best-performing model, Claude-4.5-Sonnet, achieves only a 38.6% success rate with 20.2 tool calling turns on average, while the top open-weights model DeepSeek-V3.2-Exp reaches 20.1%. We expect Toolathlon to drive the development of more capable language agents for real-world, long-horizon task execution.

Recurrent neural networks (RNNs) provide a powerful approach in neuroscience to infer latent dynamics in neural populations and to generate hypotheses about the neural computations underlying behavior. However, past work has focused on relatively simple, input-driven, and largely deterministic behaviors - little is known about the mechanisms that would allow RNNs to generate the richer, spontaneous, and potentially stochastic behaviors observed in natural settings. Modeling with Hidden Markov Models (HMMs) has revealed a segmentation of natural behaviors into discrete latent states with stochastic transitions between them, a type of dynamics that may appear at odds with the continuous state spaces implemented by RNNs. Here we first show that RNNs can replicate HMM emission statistics and then reverse-engineer the trained networks to uncover the mechanisms they implement. In the absence of inputs, the activity of trained RNNs collapses towards a single fixed point. When driven by stochastic input, trajectories instead exhibit noise-sustained dynamics along closed orbits. Rotation along these orbits modulates the emission probabilities and is governed by transitions between regions of slow, noise-driven dynamics connected by fast, deterministic transitions. The trained RNNs develop highly structured connectivity, with a small set of "kick neurons" initiating transitions between these regions. This mechanism emerges during training as the network shifts into a regime of stochastic resonance, enabling it to perform probabilistic computations. Analyses across multiple HMM architectures - fully connected, cyclic, and linear-chain - reveal that this solution generalizes through the modular reuse of the same dynamical motif, suggesting a compositional principle by which RNNs can emulate complex discrete latent dynamics.

Regression tasks, notably in safety-critical domains, require proper uncertainty quantification, yet the literature remains largely classification-focused. In this light, we introduce a family of measures for total, aleatoric, and epistemic uncertainty based on proper scoring rules, with a particular emphasis on kernel scores. The framework unifies several well-known measures and provides a principled recipe for designing new ones whose behavior, such as tail sensitivity, robustness, and out-of-distribution responsiveness, is governed by the choice of kernel. We prove explicit correspondences between kernel-score characteristics and downstream behavior, yielding concrete design guidelines for task-specific measures. Extensive experiments demonstrate that these measures are effective in downstream tasks and reveal clear trade-offs among instantiations, including robustness and out-of-distribution detection performance.

Agentic AI represents a transformative shift in artificial intelligence, but its rapid advancement has led to a fragmented understanding, often conflating modern neural systems with outdated symbolic models -- a practice known as conceptual retrofitting. This survey cuts through this confusion by introducing a novel dual-paradigm framework that categorizes agentic systems into two distinct lineages: the Symbolic/Classical (relying on algorithmic planning and persistent state) and the Neural/Generative (leveraging stochastic generation and prompt-driven orchestration). Through a systematic PRISMA-based review of 90 studies (2018--2025), we provide a comprehensive analysis structured around this framework across three dimensions: (1) the theoretical foundations and architectural principles defining each paradigm; (2) domain-specific implementations in healthcare, finance, and robotics, demonstrating how application constraints dictate paradigm selection; and (3) paradigm-specific ethical and governance challenges, revealing divergent risks and mitigation strategies. Our analysis reveals that the choice of paradigm is strategic: symbolic systems dominate safety-critical domains (e.g., healthcare), while neural systems prevail in adaptive, data-rich environments (e.g., finance). Furthermore, we identify critical research gaps, including a significant deficit in governance models for symbolic systems and a pressing need for hybrid neuro-symbolic architectures. The findings culminate in a strategic roadmap arguing that the future of Agentic AI lies not in the dominance of one paradigm, but in their intentional integration to create systems that are both adaptable and reliable. This work provides the essential conceptual toolkit to guide future research, development, and policy toward robust and trustworthy hybrid intelligent systems.

Robotic manipulation of deformable and fragile objects presents significant challenges, as excessive stress can lead to irreversible damage to the object. While existing solutions rely on accurate object models or specialized sensors and grippers, this adds complexity and often lacks generalization. To address this problem, we present a vision-based reinforcement learning approach that incorporates a stress-penalized reward to discourage damage to the object explicitly. In addition, to bootstrap learning, we incorporate offline demonstrations as well as a designed curriculum progressing from rigid proxies to deformables. We evaluate the proposed method in both simulated and real-world scenarios, showing that the policy learned in simulation can be transferred to the real world in a zero-shot manner, performing tasks such as picking up and pushing tofu. Our results show that the learned policies exhibit a damage-aware, gentle manipulation behavior, demonstrating their effectiveness by decreasing the stress applied to fragile objects by 36.5% while achieving the task goals, compared to vanilla RL policies.

Multi-agent reinforcement learning (MARl) has achieved strong results in cooperative tasks but typically assumes fixed, fully controlled teams. Ad hoc teamwork (AHT) relaxes this by allowing collaboration with unknown partners, yet existing variants still presume shared conventions. We introduce Multil-party Ad Hoc Teamwork (MAHT), where controlled agents must coordinate with multiple mutually unfamiliar groups of uncontrolled teammates. To address this, we propose MARs, which builds a sparse skeleton graph and applies relational modeling to capture cross-group dvnamics. Experiments on MPE and starCralt ll show that MARs outperforms MARL and AHT baselines while converging faster.

Bayesian inference of Bayesian network structures is often performed by sampling directed acyclic graphs along an appropriately constructed Markov chain. We present two techniques to improve sampling. First, we give an efficient implementation of basic moves, which add, delete, or reverse a single arc. Second, we expedite summing over parent sets, an expensive task required for more sophisticated moves: we devise a preprocessing method to prune possible parent sets so as to approximately preserve the sums. Our empirical study shows that our techniques can yield substantial efficiency gains compared to previous methods.

Causal discovery aims to recover graphs that represent causal relations among given variables from observations, and new methods are constantly being proposed. Increasingly, the community raises questions about how much progress is made, because properly evaluating discovered graphs remains notoriously difficult, particularly under latent confounding. We propose a graph distance measure for acyclic directed mixed graphs (AD-MGs) based on the downstream task of cause-effect estimation under unobserved confounding. Our approach uses identification via fixing and a symbolic verifier to quantify how graph differences distort cause-effect esti-mands for different treatment-outcome pairs. We analyze the behavior of the measure under different graph perturbations and compare it against existing distance metrics.

Modeling complex, non-linear, and uncertain relationships between input and output variables remains a central challenge in modern statistical learning and artificial intelligence. Traditional neural networks, trained via point estimation, have demonstrated remarkable success in a variety of domains but inherently provide deterministic predictions - that is, single-valued outputs without accompanying measures of uncertainty. This limitation becomes critical in domains characterized by limited, noisy, or ambiguous data, such as medicine, climate science, or finance, where quantifying uncertainty is as important as producing accurate predictions (Gal & Ghahramani, 2016; Kendall & Gal, 2017; Abdar et al., 2021). Bayesian Neural Networks (BNNs) provide a probabilistic extension of standard neural networks by treating weights and biases as random variables endowed with prior distributions (MacKay, 1992; Neal, 2012). Through Bayes' theorem, BNNs infer a posterior distribution over weights, allowing predictions to reflect epistemic uncertainty - the uncertainty arising from limited data and model knowledge. However, the exact posterior is analytically intractable for deep models, motivating approximate inference methods such as variational inference (Graves, 2011; Blundell et al., 2015) and Monte Carlo dropout (Gal & Ghahramani, 2016). Despite their appeal, these approaches may yield biased or overconfident posteriors due to restrictive variational families (Hern andez-Lobato & Adams, 2015a; Osband et al., 2023), often resulting in over-smoothed predictive distributions. An alternative paradigm for probabilistic modeling is the Mixture Density Network (MDN), introduced by Bridle (1990) and developed further by Jacobs et al. (1991).