arXiv subjects

Zhihuan Huang

Publications and source records attributed to Zhihuan Huang.

9 recordsLinked to original sources

arXiv2026

DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale

Large-scale agentic training and evaluation with large language models (LLMs) rely on isolated, stateful execution environments in which models inspect repositories, invoke tools, execute commands, and interact with task-specific services. These workloads create sandboxes in large bursts, span heterogeneous functionality and isolation requirements, retain state across long interactions, and draw from large image corpora with limited reuse. Supporting them therefore requires an elastic execution platform rather than a single sandbox runtime. This report presents DeepSeek Elastic Compute (DSec), a production sandbox platform that exposes FnCall, container, microVM, and full-VM sandbox backends through a unified SDK. DSec coordinates placement and lifecycle management across the cluster, composes environments from independently versioned layers, combines memory sharing, reclamation, and CPU scheduling for high-density execution, and loads image data on demand from Fire-Flyer File System (3FS), a cluster-wide distributed filesystem. DSec is co-designed with the reinforcement learning (RL) framework, decouples stateful rollout execution from preemptible GPU training, coordinates sandbox lifecycle with training to preserve rollout state while reclaiming idle resources, and mitigates agent misbehavior such as reward hacking. A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second. Our evaluation and deployment experience show that these mechanisms reduce environment setup and image-distribution overhead, improve memory efficiency, and preserve latency-sensitive performance under high-density overcommit.

cs.DC↗

arXiv2026

PMIScore: An Unsupervised Approach to Quantify Dialogue Engagement

High dialogue engagement is a crucial indicator of an effective conversation. A reliable measure of engagement could help benchmark large language models, enhance the effectiveness of human-computer interactions, or improve personal communication skills. However, quantifying engagement is challenging, since it is subjective and lacks a "gold standard". This paper proposes PMIScore, an efficient unsupervised approach to quantify dialogue engagement. It uses pointwise mutual information (PMI), which is the probability of generating a response conditioning on the conversation history. Thus, PMIScore offers a clear interpretation of engagement. As directly computing PMI is intractable due to the complexity of dialogues, PMIScore learned it through a dual form of divergence. The algorithm includes generating positive and negative dialogue pairs, extracting embeddings by large language models (LLMs), and training a small neural network using a mutual information loss function. We validated PMIScore on both synthetic and real-world datasets. Our results demonstrate the effectiveness of PMIScore in PMI estimation and the reasonableness of the PMI metric itself.

cs.CL↗

arXiv2026

Decision Aggregation under Quantal Response

The effectiveness of collective decision-making is often challenged by the bounded rationality and inherent stochasticity of individual agents. We investigate this by analyzing how to aggregate decisions from n experts, each receiving a private signal about an unknown state. Assuming signals are conditionally independent and identically distributed, we depart from the fully rational paradigm and model expert behavior using quantal response, a stochastic choice model capturing bounded rationality. Within a minimax regret framework, we show that majority voting is the optimal robust aggregator when individual rationality falls below a certain threshold. Interestingly, such groups can outperform perfectly rational agents, as their decision randomness encodes weak but informative signals lost in deterministic behavior. We validate these findings using large language models (LLMs), which naturally exhibit quantal response via their temperature parameter. Aggregating moderately stochastic LLM outputs significantly improves accuracy on complex reasoning tasks, highlighting bounded rationality not as a limitation, but as a potential strength in collective intelligence.

cs.GT↗

arXiv2024

How Gold to Make the Golden Snitch: Designing the "Game Changer" in Esports

Many battling games utilize a special item (e.g. Roshan in Defense of the Ancients 2 (DOTA 2), Baron Nashor in League of Legends (LOL), Golden Snitch in Quidditch) as a potential ``Game Changer''. The reward of this item can enable the underdog to make a comeback. However, if the reward is excessively high, the whole game may devolve into a chase for the ``Game Changer''. Our research initiates with a Quidditch case study, a fictional sport in Harry Potter series, wherein we architect the Golden Snitch's reward to maximize the audience's surprise. Surprisingly, we discover that for equally competent teams, the optimal Snitch reward is zero. Moreover, we establish that under most circumstances the optimal score aligns with the game's expected duration multiplied by the teams' strength difference. Finally, we explore the correlation between the ``Game Changer's'' reward and audience surprise in Multiplayer Online Battle Arena (MOBA) games including DOTA 2 and LOL, finding that the optimal reward escalates with increasing team strength inequality.

cs.GT↗

arXiv2024

Algorithmic Robust Forecast Aggregation

Forecast aggregation combines the predictions of multiple forecasters to improve accuracy. However, the lack of knowledge about forecasters' information structure hinders optimal aggregation. Given a family of information structures, robust forecast aggregation aims to find the aggregator with minimal worst-case regret compared to the omniscient aggregator. Previous approaches for robust forecast aggregation rely on heuristic observations and parameter tuning. We propose an algorithmic framework for robust forecast aggregation. Our framework provides efficient approximation schemes for general information aggregation with a finite family of possible information structures. In the setting considered by Arieli et al. (2018) where two agents receive independent signals conditioned on a binary state, our framework also provides efficient approximation schemes by imposing Lipschitz conditions on the aggregator or discrete conditions on agents' reports. Numerical experiments demonstrate the effectiveness of our method by providing a nearly optimal aggregator in the setting considered by Arieli et al. (2018).

cs.LG↗

arXiv2022

Eliciting Thinking Hierarchy without a Prior

When we use the wisdom of the crowds, we usually rank the answers according to their popularity, especially when we cannot verify the answers. However, this can be very dangerous when the majority make systematic mistakes. A fundamental question arises: can we build a hierarchy among the answers \textit{without any prior} where the higher-ranking answers, which may not be supported by the majority, are from more sophisticated people? To address the question, we propose 1) a novel model to describe people's thinking hierarchy; 2) two algorithms to learn the thinking hierarchy without any prior; 3) a novel open-response based crowdsourcing approach based on the above theoretic framework. In addition to theoretic justifications, we conduct four empirical crowdsourcing studies and show that a) the accuracy of the top-ranking answers learned by our approach is much higher than that of plurality voting (In one question, the plurality answer is supported by 74 respondents but the correct answer is only supported by 3 respondents. Our approach ranks the correct answer the highest without any prior); b) our model has a high goodness-of-fit, especially for the questions where our top-ranking answer is correct. To the best of our knowledge, we are the first to propose a thinking hierarchy model with empirical validations in the general problem-solving scenarios; and the first to propose a practical open-response based crowdsourcing approach that beats plurality voting without any prior.

cs.GT↗

arXiv2022

Modularity and Mutual Information in Networks: Two Sides of the Same Coin

Modularity, first proposed by [Newman and Girvan, 2004], is one of the most popular ways to quantify the significance of community structure in complex networks. It can serve as both a standard benchmark to compare different community detection algorithms, and an optimization objective to detect communities itself. Previous work on modularity has developed many efficient algorithms for modularity maximization. However, few of researchers considered the interpretation of the modularity function itself. In this paper, we study modularity from an information-theoretical perspective and show that modularity and mutual information in networks are essentially the same. The main contribution is that we develop a family of generalized modularity measures, f-modularity based on f-mutual information. f-Modularity has an information-theoretical interpretation, enjoys the desired properties of mutual information measure, and provides an approach to estimate the mutual information between discrete random variables. At a high level, we show the significance of community structure is equivalent to the amount of information contained in the network. The connection of f-modularity and f-mutual information bridges two important fields, complex network and information theory and also sheds light on the design of measures on community structure in future.

cs.SI↗

arXiv2021

BONUS! Maximizing Surprise

Multi-round competitions often double or triple the points awarded in the final round, calling it a bonus, to maximize spectators' excitement. In a two-player competition with $n$ rounds, we aim to derive the optimal bonus size to maximize the audience's overall expected surprise (as defined in [7]). We model the audience's prior belief over the two players' ability levels as a beta distribution. Using a novel analysis that clarifies and simplifies the computation, we find that the optimal bonus depends greatly upon the prior belief and obtain solutions of various forms for both the case of a finite number of rounds and the asymptotic case. In an interesting special case, we show that the optimal bonus approximately and asymptotically equals to the "expected lead", the number of points the weaker player will need to come back in expectation. Moreover, we observe that priors with a higher skewness lead to a higher optimal bonus size, and in the symmetric case, priors with a higher uncertainty also lead to a higher optimal bonus size. This matches our intuition since a highly asymmetric prior leads to a high "expected lead", and a highly uncertain symmetric prior often leads to a lopsided game, which again benefits from a larger bonus.

cs.GT↗

arXiv2021

SURPRISE! and When to Schedule It

Information flow measures, over the duration of a game, the audience's belief of who will win, and thus can reflect the amount of surprise in a game. To quantify the relationship between information flow and audiences' perceived quality, we conduct a case study where subjects watch one of the world's biggest esports events, LOL S10. In addition to eliciting information flow, we also ask subjects to report their rating for each game. We find that the amount of surprise in the end of the game plays a dominant role in predicting the rating. This suggests the importance of incorporating when the surprise occurs, in addition to the amount of surprise, in perceived quality models. For content providers, it implies that everything else being equal, it is better for twists to be more likely to happen toward the end of a show rather than uniformly throughout.

cs.MA↗