Publications

Inference for subgraph densities in noisy dynamic networks
Peter W. MacDonald
Eric D. Kolaczyk
In this work we develop statistical methodology to estimate and perform inference on subgraph densities using time-indexed, or dynamic netwo… (see more)rk sequences. These estimates explicitly adjust for observation errors for the network edges, and have good theoretical properties as the size of the network grows. By specifying a stochastically evolving hidden Markov network model, we address two important directions for further investigation identified by Chang et al. (2022): robustness to non-identical network replicates, and efficient aggregation of multiple available network snapshots. These new methods vastly expand the analysis of noisy networks to new data settings, as network replicates are commonly observed dynamically. The methodology is also extended to consider joint inference for subgraph densities at multiple time points, to facilitate formal statistical comparison of dynamic network snapshots.
On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
Andrei Liviu Nicolicioiu
On-policy self-distillation achieves strong pass@1 accuracy by using a single model as both teacher and student, with the teacher conditione… (see more)d on a correct demonstration to provide dense token-level feedback. We show that this could come at a hidden cost: rollout diversity decreases and pass@k curves flatten (i.e., generating more rollouts fails to improve accuracy). We trace this to compounding biases in the design of self-distillation with sampled demonstrations. The teacher scores each student rollout while conditioned on a sampled correct rollout, channeling its feedback through the model's own biases. We theoretically analyze the optimal self-distillation policy and show that it tilts the base distribution by a pointwise conditional mutual information score between the student's rollout and the correct rollout used as context. Unlike the ideal optimal on-policy reinforcement learning (RL), which preserves probability ratios among equally correct rollouts, self-distillation can amplify existing probability gaps, concentrating mass on already-dominant modes. On a controlled graph path-finding task and science question-answering benchmarks, self-distilled models match or exceed RL on average performance but exhibit substantially lower \textit{functional} and \textit{semantic} diversity, failing on out-of-distribution settings that require diverse strategies.
Dynamically Allocating Evaluation Effort for Model Ranking
Vilém Zouhar
Alon Lavie
Tom Kocmi
Matt Post
Ondrej Bojar
Mrinmaya Sachan
While human evaluation is the gold standard in many NLP tasks, it suffers from prohibitive costs and poor scalability. When identifying top-… (see more)performing models, typical evaluation protocols waste effort by exhaustively evaluating all models on the entire benchmark, a safe but inefficient approach. In this work, we formalize multi-model human evaluation as a best-arm identification problem in a multi-armed bandit setup with correlated arms, where pulling an arm corresponds to human-evaluating a model. By sampling adaptively based on the intermediate model rankings obtained on the samples so far, we can focus the annotation budget on the most competitive models. We prove the optimality of the proposed algorithms and show that it improves discrimination between top-performing models. This makes evaluations faster, cheaper and more aligned with large-scale competition evaluation goals.
A game theory for foundation models shows new paths to rational cooperation through similarity inference
Alexander Meulemans
Maciej Wołczyk
Marissa A. Weis
Rajai Nasser
Roberta Rocca
Seijin Kobayashi
Angelika Steger
Marcus Hutter
James Manyika
Rif A. Saurous
João Sacramento
Blaise Agüera y Arcas
As autonomous agents powered by foundation models are increasingly integrated into social and economic systems, understanding the principles… (see more) governing their collective behavior is essential for ensuring safety and cooperation. Classical game theory, the dominant framework for modeling rational interaction, is built upon the assumption of `decoupled agency,'where agents treat their own decision-making as independent of the environment and other actors. Modern AI agents, however, jointly predict their own future actions alongside external observations. Here, we report a striking finding: when interacting in stylized social dilemmas, foundation model agents engaging in optimal planning consistently converge to stable cooperation, directly contradicting classical game-theoretic predictions of mutual defection. To understand this phenomenon, we introduce the `embedded Bayesian agent,'a theoretical model for foundation model agents. By shifting from decoupled to embedded agency, these agents model themselves as part of the universe they inhabit, maintaining epistemic uncertainty about their own decision-making algorithms. We show that by inferring whether others are behaviorally similar, an embedded agent treats its own deliberation during planning as evidence: a decision to cooperate predicts a similar decision by a similar partner. We formalize this mechanism of similarity inference through the `embedded equilibrium,'a novel solution concept replacing the Nash equilibrium to provide a foundational game theory for the social behavior of modern AI agents.
Learning, Fast and Slow: Towards LLMs That Adapt Continually
Rishabh Tiwari
Lakshya A Agrawal
Joseph E. Gonzalez
Matei Zaharia
Kurt Keutzer
Inderjit S Dhillon
Devvrit Khatri
Large language models (LLMs) are trained for downstream tasks by updating their parameters (e.g., via RL). However, updating parameters forc… (see more)es them to absorb task-specific information, which can result in catastrophic forgetting and loss of plasticity. In contrast, in-context learning with fixed LLM parameters can cheaply and rapidly adapt to task-specific requirements (e.g., prompt optimization), but cannot by itself typically match the performance gains available through updating LLM parameters. There is no good reason for restricting learning to being in-context or in-weights. Moreover, humans also likely learn at different time scales (e.g., System 1 vs 2). To this end, we introduce a fast-slow learning framework for LLMs, with model parameters as "slow" weights and optimized context as "fast" weights. These fast "weights" can learn from textual feedback to absorb the task-specific information, while allowing slow weights to stay closer to the base model and persist general reasoning behaviors. Fast-Slow Training (FST) is up to 3x more sample-efficient than only slow learning (RL) across reasoning tasks, while consistently reaching a higher performance asymptote. Moreover, FST-trained models remain closer to the base LLM (up to 70% less KL divergence), resulting in less catastrophic forgetting than RL-training. This reduced drift also preserves plasticity: after training on one task, FST trained models adapt more effectively to a subsequent task than parameter-only trained models. In continual learning scenarios, where task domains change on the fly, FST continues to acquire each new task while parameter-only RL stalls.
LLM Serving in the Wild: An Empirical Study of Frameworks, Methods, and System Designs
Mohammad Mehdi Morovati
Heng Li
Large Language Models (LLMs) are integrated into software systems and AI services, making efficient LLM serving a concern for software engin… (see more)eering. Serving LLMs is challenging because inference requires computation, memory, GPU resources, and execution while maintaining latency and throughput. Although prior research has proposed LLM inference, optimization, and serving techniques and frameworks, little is known about how they are adopted in practice. In this study, we investigate the use of LLM serving frameworks and serving methods in open-source software systems. We identify and analyze five LLM-specific frameworks: vLLM, SGLang, TensorRT-LLM, LMDeploy, and FlashInfer. We examine how these frameworks and techniques are adopted individually and in combination, how adoption varies across categories of LLMs, and how repositories differ in intent, focus, use case, and architectural design. Our results show that vLLM is the most visible framework in popularity and adoption, while parallel computation, memory management, and network pruning are the most frequently used serving-method categories. Multi-framework usage is limited, suggesting that developers rely on a single serving framework; however, combined frameworks connect complementary capabilities across the serving stack. Framework adoption varies across model families, modalities, model sizes, domain specializations, and deployment settings. Repository-level analysis shows that LLM serving frameworks support applications and architectures, including Reinforcement Learning (RL)-based reasoning, multimodal generation and understanding, microservices, and cloud infrastructure. Overall, this study provides a large-scale empirical characterization of LLM serving framework adoption in practice and offers insights for researchers, framework maintainers, and practitioners working on LLM systems.
Merging Adapted Models via Data-Free Covariance Estimation
Derek Tam
Pascal Junior Tikeng Notsawo
Colin Raffel
Model merging provides a way of cheaply combining individual models to produce a model that inherits each individual's capabilities. While s… (see more)ome merging methods can approach the performance of multitask training, they are often heuristically motivated and lack theoretical justification. A principled alternative is to pose model merging as a layer-wise optimization problem that directly minimizes interference between tasks. However, this formulation requires estimating per-layer covariance matrices from data, which may not be available when performing merging. In contrast, many of the heuristically-motivated methods do not require auxiliary data, making them practically advantageous. In this work, we revisit the interference minimization framework and show that, under certain conditions, covariance matrices can be estimated directly from difference matrices, eliminating the need for data while also reducing computational costs. We validate our approach across vision and language benchmarks on models ranging from
A Survey of Agent Memory in the Second Half: Towards Self-Evolving and Long-Horizon Agents
Wei-Chieh Huang
Weizhi Zhang
Yueqing Liang
Yuanchen Bei
Yankai Chen
Tao Feng
xinyu Pan
Zhen Tan
Yu Wang
Tianxin Wei
Shanglin Wu
Ruiyao Xu
Liangwei Yang
Rui Yang
Wooseong Yang
Chin-Yuan Yeh
Hanrong Zhang
Haozhen Zhang
Siqi Zhu
Henry Peng Zou … (see 40 more)
Wanjia Zhao
Song Wang
Wujiang Xu
Zixuan Ke
Zheng Hui
Dawei Li
Yaozu Wu
Langzhou He
Chen Wang
Xiongxiao Xu
Baixiang Huang
Juntao Tan
Shelby Heinecke
Huan Wang
Caiming Xiong
Ahmed Metwally
Jun Yan
Chen-Yu Lee
Hanqing Zeng
Yinglong Xia
Xiaokai Wei
Ali Payani
Yu Wang
Haitong Ma
Wenya Wang
Chenguang Wang
Yu Zhang
Xin Eric Wang
Yongfeng Zhang
Jiaxuan You
Hanghang Tong
Xiao Luo
Xue Liu
Yizhou Sun
Wei Wang
Julian McAuley
James Zou
Jiawei Han
Philip S. Yu
Kai Shu
Research in artificial intelligence is undergoing a paradigm shift from prioritizing model innovations and benchmark scores towards emphasiz… (see more)ing problem definition and rigorous real-world evaluation. As the field enters the "second half," the central challenge becomes real utility in long-horizon, dynamic, and user-dependent settings such as agentic coding, deep research, and computer use, where LLM-based agents face context explosion beyond fixed context windows and must continuously accumulate, manage, and selectively reuse large volumes of information across extended interactions. Memory, with hundreds of papers released in 2025, therefore emerges as the critical solution to fill the utility gap. Beyond serving as passive storage, memory is increasingly the substrate through which agents self-evolve: short-term memory gates which experiences are perceived, selected, and abstracted during execution, while long-term memory accumulates and consolidates them into reusable knowledge and skills, forming the loop through which agents improve from their own experience and sustain continual learning. In this survey, we provide a unified view of foundation agent memory along three dimensions: memory substrate (internal parametric state and external retrieval-augmented stores), cognitive mechanism (sensory, working, episodic, semantic, and procedural), and memory subject (user-centric personalization and agent-centric experience). We then analyze how memory is operated under single- and multi-agent topologies and highlight learning policies over memory operations, showing how memory management itself is becoming a trainable, self-evolving capability that spans reinforcement-learned context curation, experience consolidation at decision time, and the emerging ecosystem of explicit, portable, and shareable agent skills surfaced through agent harnesses, context engineering, and standardized tool-mediation protocols. Finally, we review evaluation benchmarks and metrics for assessing memory utility, and outline various open challenges and future directions.
Analytic Planning under Uncertainty with Moment Closure
Effective model-based reinforcement learning in stochastic environments requires planning that accounts for predictive uncertainty. Propagat… (see more)ing full state distributions analytically offers a principled way to do this, but has traditionally required restrictive policy or reward structures to remain tractable. Consequently, modern deep reinforcement learning has largely retreated to either stochastic sampling, which introduces significant target variance, or deterministic point estimates that ignore predictive covariance entirely. We investigate whether distribution-aware planning is possible without these constraints. Using a quadratic action-value parameterization, we first reduce the Bellman backup to an expectation over the state-value function alone; the key idea is then a compatibility principle between the predictive transition distribution and the value function class, under which this expectation is analytic in the distribution's moments. We instantiate this principle with a Gaussian transition model paired with a radial-basis value function, yielding a closed-form backup that propagates both predictive mean and covariance. Empirically, our approach reduces target variance and yields well-calibrated predictive uncertainty under stochastic observations in continuous control, providing a principled framework for planning with learned distribution models.
Can LLMs Faithfully Enact Superforecaster Personas?
Andrew Robert Williams
Evan Jiang
Nasim Rahaman
Superforecasters are people that make forecasts that are statistically better than the average person; they offer an interesting testbed for… (see more) persona modelling. In this preliminary work, we investigate whether large language models (LLMs) can faithfully model superforecaster-like personas making similar forecasts and generating similar rationales behind the forecasts. When we compare LLM-generated responses to the ground truth responses of superforecasters, we find that few-shot prompting generates numerically closer forecasts than instruction-prompting baselines. We also measure the similarity of generated and real rationales, finding that the rationales generated based on examples are judged more similar to those of superforecasters than instruction-prompting baselines. However, an analysis of the reasoning patterns in the rationales shows significant differences between human and LLM forecasters, pointing to a gap between stylistic imitation and deep reasoning similarity. These preliminary results raise several interesting questions about how getting superforecaster-like behaviour from LLMs actually works, and open new avenues to explore for improving the forecasting behaviour of LLMs.
Real-time tissue-equivalent measurement of individual clinical radiotherapy pulses
Fernanda C. Rodrigues-Machado
Katherine Szabo
Jingyi Bian
Simon Bernard
Tanner Connell
Shirin Abbasinejad Enger
Lilian Childress
Jack C. Sankey
We apply the precision tools of cavity-enhanced absorption sensing to clinical oncology, demonstrating a dosimeter paradigm in which a centi… (see more)meter-scale volume of water serves as a tissue-equivalent sensing medium. Our proof-of-concept, all-optical scheme achieves real-time readout of clinical radiation pulses with a nominal single-pulse resolution of 90
Speculative Successive Cancellation Decoding of Polar Codes
Ryan Seah
Marvin Rübenacke
Warren J. Gross
Polar codes achieve channel capacity as block length increases, but this asymptotic advantage comes at the cost of decoding speed: the conve… (see more)ntional successive cancellation (SC) algorithm is inherently sequential, which limits its practicality for high-throughput applications. This work addresses that limitation by introducing the \emph{Speculative Successive-Cancellation} (Spec-SC) framework, which incorporates speculative execution into the SC decoding process. While node-based decoders with special nodes enable partial parallelism for specific, predefined structures, Spec-SC extends this capability by permitting parallel execution in general nodes that do not match any special-node pattern. A built-in verification mechanism ensures that this speculation does not compromise decoding correctness. Simulation results show that Spec-SC achieves identical BER and FER performance to standard SC decoding, while reducing the average number of sequentially traversed nodes by up to