arXiv subjects

Kefei Chen

Publications and source records attributed to Kefei Chen.

11 recordsLinked to original sources

arXiv2026

AgentRewind: Recoverable Execution for Long-Horizon LLM Agents

Many real-world tasks require LLM agents to interact with their environments over long execution horizons. Errors that occur early in execution may propagate through both the agent context and environment state, and their effects may be difficult to reverse through subsequent actions. Existing methods mainly seek to reduce such errors through plan refinement and safety checks but provide little support after errors occur. To enable recovery during long-horizon execution, we present AgentRewind, a runtime recovery framework that records aligned checkpoints of the agent context and controlled environment, allowing agents to return to an earlier state and resume execution with information from previous attempts. We also construct MettleBench, a benchmark for evaluating task completion and partial progress on long-horizon engineering assignments containing a series of related requirements. Experiments across tasks, multiple models, execution strategies, and agent harnesses show that AgentRewind improves task success rate and average checklist progress over the compared baselines.

cs.AI↗

arXiv2026

FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards

Live future prediction refers to the task of making predictions about real-world events before they unfold. This task is increasingly studied using large language model-based agent systems, and it is important for building agents that can continually learn from the real world. It can provide a large number of prediction questions grounded in diverse real-world events, while preventing answer leakage. To leverage the advantages of future prediction, we present FutureWorld, a live agentic reinforcement learning environment that closes the training loop between prediction, outcome realization, and parameter updates. Specifically, we modify and extend verl-tool, resulting in a new framework that we call verl-tool-future. Unlike standard reinforcement learning training frameworks that rely on immediate rewards, verl-tool-future stores prediction-time rollouts, backfills rewards after real-world outcomes become available, and then replays the completed trajectories for policy update. Across three open-source agents, successive FutureWorld training rounds lead to consistent improvements in prediction accuracy, probabilistic scoring, and calibration, demonstrating that delayed real-world outcome feedback can serve as an effective reinforcement learning signal.

cs.AI↗

arXiv2026

Harnessing Pre-Resolution Signals for Future Prediction Agents

Many high-stakes decisions depend on forecasts made before outcomes are known. In this future prediction setting, the central challenge is that public evidence evolves over time, while the main supervision signal arrives only after resolution: the realized outcome mainly assesses final correctness, offering only coarse guidance on what to track, what to verify, and which judgments to leave uncertain along the way. Our key observation is that revisiting the same unresolved question over time creates informative temporal contrasts across evolving evidence and repeated forecasts, exposing what earlier attempts missed before resolution and yielding a diagnostic signal we call the pre-resolution signal. We instantiate this idea in Milkyway, a future prediction agent with a persistent future prediction harness, an editable external state that stores reusable procedural guidance across revisits to the same unresolved question. As the same unresolved question is revisited, Milkyway extracts pre-resolution signals from evolving evidence and repeated forecasts, uses them to update the harness, and improves later forecasts on that question before resolution. After resolution, the realized outcome serves as a post-resolution check of provisional updates. On the FutureX and FutureWorld benchmarks, Milkyway achieves strong performance against competitive baselines, and a mechanism study suggests that the gains stem from harness evolution driven by pre-resolution signals rather than repeated prediction alone.

cs.AI↗

arXiv2019

A framework for quantum homomorphic encryption with experimental demonstration

Quantum homomorphic encryption (QHE) is an encryption method that allows quantum computation to be performed on one party's private data with the program provided by another party, without revealing much information about the data nor the program to the opposite party. We propose a framework for (interactive) QHE based on the universal circuit approach. It contains a subprocedure of calculating a classical linear polynomial, which can be implemented with quantum or classical methods; apart from the subprocedure, the framework has low requirement on the quantum capabilities of the party who provides the circuit. We illustrate the subprocedure using a quite simple classical protocol with some privacy tradeoff. For a special case of such protocol, we obtain a scheme similar to blind quantum computation but with the output on a different party. Another way of implementing the subprocedure is to use a recently studied quantum check-based protocol, which has low requirement on the quantum capabilities of both parties. The subprocedure could also be implemented with a classical additive homomorphic encryption scheme. We demonstrate some key steps of the outer part of the framework in a quantum optics experiment.

quant-ph↗

arXiv2018

Experimental demonstration of quantum walks with initial superposition states

The preparation of initial superposition states of discrete-time quantum walks (DTQWs) are necessary for the study and applications of DTQWs. In linear optics, it is easy to prepare initial superposition states of the coin, which are always encoded by polarization states; while the preparation of superposition states of the walker is challenging. Based on a novel encoding method, we here propose a DTQW protocol in linear optics which enables the preparation of arbitrary initial superposition states of the walker and the coin. With this protocol, we report an experimental demonstration of DTQW with the walker initially in superposition states, by using only passive linear-optical elements. The effects of the walker's different initial superposition states on the spread speed of the DTQW and on the entanglement between the coin and the walker are also experimentally investigated, which have not been reported before. When the walker starts with superposition states, we show that the properties of DTQW are very different from those of DTQW starting with a single position. Our findings reveal different properties of DTQW and paves an avenue to study DTQW with arbitrary initial states. Moreover, the encoding method enables one to encode an arbitrary high-dimensional quantum state using a single physical qubit and may be adopted to implement other quantum information tasks.

quant-ph↗

arXiv2017

Experimental simulation of quantum temporal steering beyond rotating-wave approximation

Characterizing the dynamics of open systems usually starts with a perturbative theory and involves various approximations, such as the Born, Markov and rotating-wave approximation (RWA). However, the approximation approaches could introduce more or less incompleteness in describing the bath behaviors. Here, we consider a quantum channel, which is modeled by a qubit (a two-level system) interacting with a bosonic bath. Unlike the traditional works, we experimentally simulate the system-bath interaction without applying the Born, Markov, and rotating-wave approximations. To our knowledge, this is the first experimental simulation of the quantum channels without any approximations mentioned above, by using linear optical devices. The results are quite useful and interesting, which not only reveal the effect of the counter-rotating terms but also present a more accurate picture of the quantum channel dynamics. Besides, we experimentally investigate the dynamics of the quantum temporal steering (TS), i.e., a temporal analogue of Einstein-Podolsky-Rosen steering. The experimental and theoretical results are in good agreement and show that the counter-rotating terms significantly influence the TS dynamics. When one monogamously associates TS with the security of the cryptographic protocols (e.g., BB84), our experimental tests reveal that the channels based on RWA will provide exaggerated security durations, while they are actually insecure in non-RWA channel cases. This implies that doing RWA may result in a risk for the security of quantum key distribution. Our findings are expected to have useful applications in secure quantum communications and future interesting TS studies.

quant-ph↗

arXiv2008

MOHCS: Towards Mining Overlapping Highly Connected Subgraphs

Many networks in real-life typically contain parts in which some nodes are more highly connected to each other than the other nodes of the network. The collection of such nodes are usually called clusters, communities, cohesive groups or modules. In graph terminology, it is called highly connected graph. In this paper, we first prove some properties related to highly connected graph. Based on these properties, we then redefine the highly connected subgraph which results in an algorithm that determines whether a given graph is highly connected in linear time. Then we present a computationally efficient algorithm, called MOHCS, for mining overlapping highly connected subgraphs. We have evaluated experimentally the performance of MOHCS using real and synthetic data sets from computer-generated graph and yeast protein network. Our results show that MOHCS is effective and reliable in finding overlapping highly connected subgraphs. Keywords-component; Highly connected subgraph, clustering algorithms, minimum cut, minimum degree

cs.DC↗

arXiv2008

New Upper Bounds on Sizes of Permutation Arrays

A permutation array(or code) of length $n$ and distance $d$, denoted by $(n,d)$ PA, is a set of permutations $C$ from some fixed set of $n$ elements such that the Hamming distance between distinct members $\mathbf{x},\mathbf{y}\in C$ is at least $d$. Let $P(n,d)$ denote the maximum size of an $(n,d)$ PA. New upper bounds on $P(n,d)$ are given. For constant $α,β$ satisfying certain conditions, whenever $d=βn^α$, the new upper bounds are asymptotically better than the previous ones.

cs.IT↗

arXiv2008

New Lower Bounds on Sizes of Permutation Arrays

A permutation array(or code) of length $n$ and distance $d$, denoted by $(n,d)$ PA, is a set of permutations $C$ from some fixed set of $n$ elements such that the Hamming distance between distinct members $\mathbf{x},\mathbf{y}\in C$ is at least $d$. Let $P(n,d)$ denote the maximum size of an $(n,d)$ PA. This correspondence focuses on the lower bound on $P(n,d)$. First we give three improvements over the Gilbert-Varshamov lower bounds on $P(n,d)$ by applying the graph theorem framework presented by Jiang and Vardy. Next we show another two new improved bounds by considering the covered balls intersections. Finally some new lower bounds for certain values of $n$ and $d$ are given.

cs.IT↗

arXiv2008

New Constructions of Permutation Arrays

A permutation array(permutation code, PA) of length $n$ and distance $d$, denoted by $(n,d)$ PA, is a set of permutations $C$ from some fixed set of $n$ elements such that the Hamming distance between distinct members $\mathbf{x},\mathbf{y}\in C$ is at least $d$. In this correspondence, we present two constructions of PA from fractional polynomials over finite field, and a construction of $(n,d)$ PA from permutation group with degree $n$ and minimal degree $d$. All these new constructions produces some new lower bounds for PA.

cs.IT↗

arXiv2005

Efficient Authenticated Encryption Schemes with Public Verifiability

An authenticated encryption scheme allows messages to be encrypted and authenticated simultaneously. In 2003, Ma and Chen proposed such a scheme with public verifiability. That is, in their scheme the receiver can efficiently prove to a third party that a message is indeed originated from a specific sender. In this paper, we first identify two security weaknesses in the Ma-Chen authenticated encryption scheme. Then, based on the Schnorr signature, we proposed an efficient and secure improved scheme such that all the desired security requirements are satisfied.

cs.CR↗