Linghua Zhang | Searcharxiv

arXiv2026

RetailBench: Evaluating Long-Horizon Autonomous Decision-Making and Strategy Stability of LLM Agents in Realistic Retail Environments

Large language model (LLM) agents have made rapid progress on short-horizon, well-scoped tasks, yet their ability to sustain coherent decisions in dynamic long-horizon environments remains uncertain. We introduce RetailBench, a data-grounded simulation benchmark for evaluating tool-using LLM agents in single-store supermarket operation. RetailBench models retail management as a partially observable decision process and is designed to support thousand-day-scale simulations. In this environment, agents must manage pricing, replenishment, supplier selection, shelf assortment, inventory aging, customer feedback, external events, and cash-flow constraints. We evaluate seven contemporary LLMs under representative agent frameworks over a 180-day evaluation horizon and compare them with a privileged oracle policy. Results show substantial variation across models: only a small subset survives the full evaluation horizon, and even the strongest LLM runs remain substantially behind the oracle policy in final net worth and sales outcomes. Behavioral analysis attributes these gaps to incomplete evidence acquisition, surface-level decision making, and the lack of a consistent long-horizon policy. RetailBench provides a controlled testbed for studying reliable autonomy in economically grounded long-horizon decision-making.

cs.AI↗

arXiv2026

RetailBench: Benchmarking long horizon reasoning and coherent decision making of LLM agents in realistic retail environments

Large language model (LLM) agents have made rapid progress on short-horizon, well-scoped tasks, yet their ability to sustain coherent decisions in dynamic long-horizon environments remains uncertain. We introduce RetailBench, a data-grounded simulation benchmark for evaluating tool-using LLM agents in single-store supermarket operation. RetailBench models retail management as a partially observable decision process and is designed to support thousand-day-scale simulations. In this environment, agents must manage pricing, replenishment, supplier selection, shelf assortment, inventory aging, customer feedback, external events, and cash-flow constraints. We evaluate seven contemporary LLMs under representative agent frameworks over a 180-day evaluation horizon and compare them with a privileged oracle policy. Results show substantial variation across models: only a small subset survives the full evaluation horizon, and even the strongest LLM runs remain substantially behind the oracle policy in final net worth and sales outcomes. Behavioral analysis attributes these gaps to incomplete evidence acquisition, surface-level decision making, and the lack of a consistent long-horizon policy. RetailBench provides a controlled testbed for studying reliable autonomy in economically grounded long-horizon decision-making.

cs.AI↗

arXiv2026

BEAST3D: Animal behavioral analysis and neural encoding from multi-view video via Gaussian splatting

Multi-view video recordings are increasingly used to capture the 3D movements of animals in experimental settings, yet extracting rich 3D representations from these recordings remains challenging. Supervised pose estimation requires extensive manual annotation, while general-purpose 3D reconstruction models trained on generic scene datasets fail on the specialized imagery and sparse-view setting of laboratory experiments. We address these limitations with BEAST3D, a self-supervised pretraining framework that learns 3D visual representations from unlabeled, calibrated multi-view video. BEAST3D uses a vision transformer to predict 3D Gaussian splats that reconstruct held-out views through differentiable rendering, while simultaneously segmenting the animal from the background. BEAST3D reconstructs 3D structure with as few as four views by conditioning directly on known camera parameters--unlike general-purpose models, which must estimate camera geometry from dense overlapping viewpoints that are seldom available in lab settings. Through comprehensive evaluation across four species, we demonstrate that BEAST3D produces rich, viewpoint-invariant features that transfer effectively to three downstream tasks: novel view synthesis, which validates the quality of the learned 3D representations; multi-view pose estimation, which provides the sparse keypoint trajectories widely used in behavioral analysis; and neural encoding, which relates 3D behavioral features to simultaneously recorded neural activity. BEAST3D thus establishes a versatile framework for behavioral analysis that leverages 3D structure in modern multi-view laboratory recordings.

q-bio.NC↗

arXiv2019

Deep Reinforcement Learning for Smart Home Energy Management

In this paper, we investigate an energy cost minimization problem for a smart home in the absence of a building thermal dynamics model with the consideration of a comfortable temperature range. Due to the existence of model uncertainty, parameter uncertainty (e.g., renewable generation output, non-shiftable power demand, outdoor temperature, and electricity price) and temporally-coupled operational constraints, it is very challenging to determine the optimal energy management strategy for scheduling Heating, Ventilation, and Air Conditioning (HVAC) systems and energy storage systems in the smart home. To address the challenge, we first formulate the above problem as a Markov decision process, and then propose an energy management strategy based on Deep Deterministic Policy Gradients (DDPG). It is worth mentioning that the proposed strategy does not require the prior knowledge of uncertain parameters and building thermal dynamics model. Simulation results based on real-world traces demonstrate the effectiveness and robustness of the proposed strategy.

eess.SY↗

arXiv2017

Power-Constrained Secrecy Rate Maximization for Joint Relay and Jammer Selection Assisted Wireless Networks

In this paper, we examine the physical layer security for cooperative wireless networks with multiple intermediate nodes, where the decode-and-forward (DF) protocol is considered. We propose a new joint relay and jammer selection (JRJS) scheme for protecting wireless communications against eavesdropping, where an intermediate node is selected as the relay for the sake of forwarding the source signal to the destination and meanwhile, the remaining intermediate nodes are employed to act as friendly jammers which broadcast the artificial noise for disturbing the eavesdropper. We further investigate the power allocation among the source, relay and friendly jammers for maximizing the secrecy rate of proposed JRJS scheme and derive a closed-form sub-optimal solution. Specificially, all the intermediate nodes which successfully decode the source signal are considered as relay candidates. For each candidate, we derive the sub-optimal closed-form power allocation solution and obtain the secrecy rate result of the corresponding JRJS scheme. Then, the candidate which is capable of achieving the highest secrecy rate is selected as the relay. Two assumptions about the channel state information (CSI), namely the full CSI (FCSI) and partial CSI (PCSI), are considered. Simulation results show that the proposed JRJS scheme outperforms the conventional pure relay selection, pure jamming and GSVD based beamforming schemes in terms of secrecy rate. Additionally, the proposed FCSI based power allocation (FCSI-PA) and PCSI based power allocation (PCSI-PA) schemes both achieve higher secrecy rates than the equal power allocation (EPA) scheme.

cs.IT↗