arXiv subjects

Takahiro Ezaki

Publications and source records attributed to Takahiro Ezaki.

At least 19 recordsLinked to original sources

arXiv2026

Shared and structured inputs undermine collective random choice by reasoning AI agents

Random selection is widely used in resource allocation and auditing, making reliable implementation essential for AI-agent systems. Behavioural tests across six reasoning models uncovered threshold and divisibility rules used in identifier-based choices. For threshold-following GPT-6 Sol and Gemini 3.8 Flash, single-agent measurements prospectively predicted correlated participation under shared identifiers and biased participation under distinct identifiers with common timestamp bits. Changing dates, formats and identifier labels revealed when these predictions held. Explicit instructions to randomize independently reduced but did not eliminate shared-input correlation. To test implications for oversight, we asked four models to select customer requests randomly for human review. GPT-6 Sol approached the target rate while selecting predictably from identifiers; the others rarely selected requests. All four closely followed supplied random draws. These findings expose collective and audit vulnerabilities that selection rates alone miss, making input-dependent bias, correlation and predictability central targets for agent evaluation.

cs.AI↗

arXiv2026

Warned alike, AI agents avoid the less-crowded road while people take it

AI agents built on a few shared models increasingly act for many people. A shared forecast about others can align their choices and change how scarce capacity is allocated. We tested this feedback in a two-road congestion game. Adding one sentence warning that others might follow a routing tip made populations of 50 GPT agents crowd one road while avoiding the nearly empty alternative. Average travel time rose from 64 to 95 min, although any crowded-road agent could have saved 69 min by switching alone. The warning discouraged the very move it predicted. The pattern persisted for 100 rounds. Two other model families shifted the same way without locking onto one road. Twelve all-human groups (240 participants) stayed near balance under numerical reports or the tip and warning. In 24 mixed groups with a further 240 participants, imbalance grew with the share of agents in the registered analysis, while people increasingly took the road the agents avoided. Collective costs stayed below the allagent reference, but with 15 agents and 5 humans, agent seats averaged 80 min, compared with 44 min for human seats. Shared forecasts can thus sustain collective inefficiency among similar agents. A better group average can also hide an unequal burden. Evaluations of AI agents that share resources should test populations, treat messages as interventions and report who bears the costs.

physics.soc-ph↗

arXiv2026

How a shared state is described determines whether AI agents synchronize

Language-model agents increasingly act in populations, where the outcome that matters is collective: whether they align, split or fail to coordinate. Each acts not on the world but on a text description of it, a choice usually fixed in software. Using synchronization, the canonical probe of how interaction rules produce collective order, we show that this choice can decide the outcome. Agents on a circle chose to advance, stay or move back after reading the others' relative positions, in 507,112 valid responses across matched populations, controlled inputs and three model families. In GPT, numerical summaries aligned every matched population at both positive couplings, whereas histograms aligned none; Claude showed the reverse at the stronger coupling. Re-describing identical states shifted action probabilities in all three families, even between histograms carrying the same information. No single directional coefficient explained the outcome: state descriptions are part of the interaction rule that turns individual responses into collective order.

physics.soc-ph↗

arXiv2026

When Shippers Become Algorithms: Candidate Exposure, Information Design, and the Concentration of LLM-Mediated Freight Markets

Shippers are beginning to delegate carrier selection to large language model (LLM) agents. We ask what such delegation does to a freight matching market, and which platform design choices contain it. We carried out agent-based simulations in which fifty shipper agents, built on commercial LLMs from OpenAI (GPT), Anthropic (Claude), and Google (Gemini), procure truckload capacity for thirty days. The market implements the rules of digital freight matching: each load is offered down the shipper's ranked list of carriers (waterfall tendering), carriers have daily capacity limits, spot prices respond to congestion, and carrier ratings accumulate with transactions. We found three risks and one remedy that works. Agents converged at once: for a fixed sampled carrier population, the same carrier was the modal first choice of every model on day one, attracting up to 76% of requests. Because each agent picks from its own randomly drawn list of displayed candidates, the platform controls how many options each shipper sees; concentration rose steeply once lists exceeded about ten carriers, with the onset differing across models. Which carriers ended up dominant varied widely from one sampled market to another, and displaying true quality instead of estimated ratings changed neither the level nor this variability (by design, quality affects only what agents see, never delivery outcomes). Against these risks, disclosing each carrier's remaining daily capacity cut concentration by a third and doubled shipper surplus, while vendor diversification, list-order randomization, and popularity display showed no clearly detectable effect. Platform information design, ahead of model choice or model regulation, is the lever that works.

physics.soc-ph↗

arXiv2022

Towards understanding network topology and robustness of logistics systems

Advanced integration of logistics systems has been promoted for the sake of competitiveness and sustainability. Such efforts will enable more globally optimal and flexible operations by efficiently utilizing transportation capacity. At the same time, interconnection of transport operations increases complexity at a network level, which reduces the predictability of the response of the system to disruptions. However, our understanding of the behavior of such systems is still limited. In particular, the topology of the network, which changes as the systems are integrated, is an important factor that affects the performance of the entire system. Knowledge of such mechanisms would be useful in the design and evaluation of integrated logistics. Here, we developed a simple mathematical model that extracts the essence of the problem and performed extensive numerical experiments by Monte Carlo simulations for three scenarios that mimic changes in demand: (i) locally and temporally increased traffic demand, (ii) globally and temporally increased traffic demand, and (iii) permanent change in demand pattern, under various conditions on the type of route-finding algorithm, network structure, and transportation capacity. Adaptive route-finding algorithms were more effective in square lattice and random networks, which contained many bypass routes, than in hub-and-spoke networks. Furthermore, the square lattice and random networks were robust to the change in the demand pattern and temporal blockage of delivery paths (e.g., due to high demand). We suggest that such preferable properties are only present in networks with redundancy and that the bypass structure is an important criterion for designing network logistics.

physics.soc-ph↗

arXiv2021

Comments on "Y.-Q. Wang, et al., Commun. Nonlinear Sci. Numer. Simul. 103:105981 (2021)"

With regard to the recently published article, ``Y.-Q. Wang, et al., Physical mechanism of equiprobable exclusion network with heterogeneous interactions in phase transitions: Analytical analyses of steady state evolving from initial state, Commun. Nonlinear Sci. Numer. Simul. 103:105981 (2021)'', the following concerns are raised. First, the given steady-state solution of the system, its proof, and derivation of relevant physical quantities are incorrect. Second, the presented simulation results were not replicated and significantly deviated from their analytical predictions. Finally, the second part of the article is irrelevant in the context of equiprobable exclusion networks because the examined system in this part does not possess equiprobable properties.

cond-mat.stat-mech↗

arXiv2021

Model retraining and information sharing in a supply chain with long-term fluctuating demands

Demand forecasting based on empirical data is a viable approach for optimizing a supply chain. However, in this approach, a model constructed from past data occasionally becomes outdated due to long-term changes in the environment, in which case the model should be updated (i.e., retrained) using the latest data. In this study, we examine the effects of updating models in a supply chain using a minimal setting. We demonstrate that when each party in the supply chain has its own forecasting model, uncoordinated model retraining causes the bullwhip effect even if a very simple replenishment policy is applied. Our results also indicate that sharing the forecasting model among the parties involved significantly reduces the bullwhip effect.

physics.soc-ph↗

arXiv2021

Modeling state-transition dynamics in resting-state brain signals by the hidden Markov and Gaussian mixture models

Recent studies have proposed that one can summarize brain activity into dynamics among a relatively small number of hidden states and that such an approach is a promising tool for revealing brain function. Hidden Markov models (HMMs) are a prevalent approach to inferring such neural dynamics among discrete brain states. However, the impact of assuming Markovian structure in neural time series data has not been sufficiently examined. Here, to address this situation and examine the performance of the HMM, we compare the model with the Gaussian mixture model (GMM), which is with no temporal regularization and thus a statistically simpler model than the HMM, by applying both models to synthetic time series generated from empirical resting-state functional magnetic resonance imaging (fMRI) data. We compared the GMM and HMM for various sampling frequencies, lengths of recording per participant, numbers of participants, and numbers of independent component signals. We find that the HMM attains a better accuracy of estimating the hidden state than the GMM in a majority of cases. However, we also find that the accuracy of the GMM is comparable to that of the HMM under the condition that the sampling frequency is reasonably low (e.g., TR = 2.88 or 3.60 s) or the data is relatively short. These results suggest that the GMM can be a viable alternative to the HMM for investigating hidden-state dynamics under this condition.

q-bio.NC↗

arXiv2018

Clustering coefficients for correlation networks

The clustering coefficient quantifies the abundance of connected triangles in a network and is a major descriptive statistics of networks. For example, it finds an application in the assessment of small-worldness of brain networks, which is affected by attentional and cognitive conditions, age, psychiatric disorders and so forth. However, it remains unclear how the clustering coefficient should be measured in a correlation-based network, which is among major representations of brain networks. In the present article, we propose clustering coefficients tailored to correlation matrices. The key idea is to use three-way partial correlation or partial mutual information to measure the strength of the association between the two neighbouring nodes of a focal node relative to the amount of pseudo-correlation expected from indirect paths between the nodes. Our method avoids the difficulties of previous applications of clustering coefficient (and other) measures in defining correlational networks, i.e., thresholding on the correlation value, discarding of negative correlation values, the pseudo-correlation problem and full partial correlation matrices whose estimation is computationally difficult. For proof of concept, we apply the proposed clustering coefficient measures to functional magnetic resonance imaging data obtained from healthy participants of various ages and compare them with conventional clustering coefficients. We show that the clustering coefficients decline with the age. The proposed clustering coefficients are more strongly correlated with age than the conventional ones are. We also show that the local variants of the proposed clustering coefficients are useful in characterising individual nodes. In contrast, the conventional local clustering coefficients were strongly correlated with and therefore may be confounded by the node's connectivity.

physics.soc-ph↗

arXiv2017

Reinforcement learning account of network reciprocity

Evolutionary game theory predicts that cooperation in social dilemma games is promoted when agents are connected as a network. However, when networks are fixed over time, humans do not necessarily show enhanced mutual cooperation. Here we show that reinforcement learning (specifically, the so-called Bush-Mosteller model) approximately explains the experimentally observed network reciprocity and the lack thereof in a parameter region spanned by the benefit-to-cost ratio and the node's degree. Thus, we significantly extend previously obtained numerical results.

physics.soc-ph↗

arXiv2017

Energy landscape analysis of neuroimaging data

Computational neuroscience models have been used for understanding neural dynamics in the brain and how they may be altered when physiological or other conditions change. We review and develop a data-driven approach to neuroimaging data called the energy landscape analysis. The methods are rooted in statistical physics theory, in particular the Ising model, also known as the (pairwise) maximum entropy model and Boltzmann machine. The methods have been applied to fitting electrophysiological data in neuroscience for a decade, but their use in neuroimaging data is still in its infancy. We first review the methods and discuss some algorithms and technical aspects. Then, we apply the methods to functional magnetic resonance imaging data recorded from healthy individuals to inspect the relationship between the accuracy of fitting, the size of the brain system to be analyzed, and the data length.

q-bio.NC↗

arXiv2016

Inflow process of pedestrians to a confined space

To better design safe and comfortable urban spaces, understanding the nature of human crowd movement is important. However, precise interactions among pedestrians are difficult to measure in the presence of their complex decision-making processes and many related factors. While extensive studies on pedestrian flow through bottlenecks and corridors have been conducted, the dominant mode of interaction in these scenarios may not be relevant in different scenarios. Here, we attempt to decipher the factors that affect human reactions to other individuals from a different perspective. We conducted experiments employing the inflow process in which pedestrians successively enter a confined area (like an elevator) and look for a temporary position. In this process, pedestrians have a wider range of options regarding their motion than in the classical scenarios; therefore, other factors might become relevant. The preference of location is visualized by pedestrian density profiles obtained from recorded pedestrian trajectories. Non-trivial patterns of space acquisition, e.g., an apparent preference for positions near corners, were observed. This indicates the relevance of psychological and anticipative factors beyond the private sphere, which have not been deeply discussed so far in the literature on pedestrian dynamics. From the results, four major factors, which we call flow avoidance, distance cost, angle cost, and boundary preference, were suggested. We confirmed that a description of decision-making based on these factors can give a rise to realistic preference patterns, using a simple mathematical model. Our findings provide new perspectives and a baseline for considering the optimization of design and safety in crowded public areas and public transport carriers.

physics.soc-ph↗

arXiv2015

Jamming transitions in force-based models for pedestrian dynamics

Force-based models describe pedestrian dynamics in analogy to classical mechanics by a system of second order ordinary differential equations. By investigating the linear stability of two main classes of forces, parameter regions with unstable homogeneous states are identified. In this unstable regime it is then checked whether phase transitions or stop-and-go waves occur. Results based on numerical simulations show, however, that the investigated models lead to unrealistic behavior in form of backwards moving pedestrians and overlapping. This is one reason why stop-and-go waves have not been observed in these models. The unrealistic behavior is not related to the numerical treatment of the dynamic equations but rather indicates an intrinsic problem of this model class. Identifying the underlying generic problems gives indications how to define models that do not show such unrealistic behavior. As an example we introduce a new force-based model which produces realistic jam dynamics without the appearance of unrealistic negative speeds for empirical desired walking speeds.

physics.soc-ph↗

arXiv2015

Taming macroscopic jamming in transportation networks

In transportation networks, a spontaneous jamming transition is often observed, e.g in urban road networks and airport networks. Because of this instability, flow distribution is significantly imbalanced on a macroscopic level. To mitigate the congestion, we consider a simple control method, in which congested nodes are closed temporarily, and investigate how it influences the overall system. Depending on the timing of the node closure and opening, and congestion level of a network, the system displays three different phases: free-flow phase, controlled phase, and deadlock phase. We show that when the system is in the controlled phase, the average flow is significantly improved, whereas when in the deadlock phase, the flow drops to zero. We study how the control method increases the network flow and obtain their transition boundary analytically.

physics.soc-ph↗

arXiv2015

Dynamics of assembly production flow

Despite recent developments in management theory, maintaining a manufacturing schedule remains difficult because of production delays and fluctuations in demand and supply of materials. The response of manufacturing systems to such disruptions to dynamic behavior has been rarely studied. To capture these responses, we investigate a process that models the assembly of parts into end products. The complete assembly process is represented by a directed tree, where the smallest parts are injected at leaves and the end products are removed at the root. A discrete assembly process, represented by a node on the network, integrates parts, which are then sent to the next downstream node as a single part. The model exhibits some intriguing phenomena, including overstock cascade, phase transition in terms of demand and supply fluctuations, nonmonotonic distribution of stockout in the network, and the formation of a stockout path and stockout chains. Surprisingly, these rich phenomena result from only the nature of distributed assembly processes. From a physical perspective, these phenomena provide insight into delay dynamics and inventory distributions in large-scale manufacturing systems.

physics.soc-ph↗

arXiv2015

Exact solution of a heterogeneous multi-lane asymmetric simple exclusion process

We prove an exact solution of a multi-lane totally asymmetric simple exclusion process (TASEP) with heterogeneous lane-changing rates on a torus. The solution is given by a factorized form; that is, the TASEP in each lane and lane-changing transition can be separable because of the detailed balance conditions in lane-changing transition. Using the saddle point method, the current of particles is calculated in a simple form in a thermodynamic limit. It is notable that the current depends only on a set of lane-changing parameters, not on the configuration of lanes.

physics.soc-ph↗

arXiv2015

Collective motion of oscillatory walkers

We study a system of interacting self-propelled particles whose walking velocity depends on the stage of the locomotion cycle. The model introduces a phase equation in the optimal velocity model for vehicular traffic. We find that the system exhibits novel types of flow: synchronized free flow, phase-anchoring free flow, orderly jam flow, and disordered jam flow. The first two flows are characterized by synchronization of the phase, while the others do not have the global synchronization. Among these, the disordered jam flow is very complex, although the underlying model is simple. This phenomenon implies that the crowd behavior of moving particles can be destabilized by coupling their velocity to the phase of their motion. We also focus on "phase-anchoring" phenomena. They strongly affect particle flow in the system, especially when the density of particles is high.

physics.soc-ph↗

arXiv2015

Potential global jamming transition in aviation networks

In this paper, we propose a nonlinear transport model for an aviation network. The takeoff rate from an airport is characterized by the degree of ground congestion. Due to the effect of "surface congestion," the performance of an airport deteriorates because of inefficient configurations of waiting aircraft on the ground. Using a simple transport model, we performed simulations on a U. S. airport network and found a global jamming transition induced by local surface congestion. From a physical perspective, the mechanism of the transition is studied analytically and the resulting aircraft distribution is discussed considering system dynamics. This study shows that the knowledge of the relationship between a takeoff rate and a congestion level on the ground is vital for efficient air traffic operations.

physics.soc-ph↗