Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks84 from public data
- 163
- 143
- EEG Cortical Connectivity Analysis of Working Memory Reveals Topological Reorganization in Theta and Alpha Bands113
Higher functional integration in the theta band and lower functional segregation in the alpha band in the WM task compared with the control task is found, which demonstrates the potential for using the brain network metrics as biomarkers for predicting the task performance during WM tasks.
- Bayesian Optimization Meets Bayesian Optimal Stopping71
This paper proposes to unify BO (specifically, Gaussian process-upper confi-dence bound (GP-UCB) with Bayesian optimal stopping (BO-BOS) with Bayesian optimal stopping (BO-BOS) to boost the epoch efficiency of BO.
- Modular‐level alterations of structure–function coupling in schizophrenia connectome67
It is demonstrated that altered module‐dependent SC–FC coupling may underlie abnormal brain function and clinical symptoms observed in schizophrenia and highlighted the potential for using new multimodal neuroimaging biomarkers for diagnosis and severity evaluation of schizophrenia.
- 51
- Prompt Optimization with Human Feedback47
The problem of prompt optimization with human feedback (POHF), in which the APOHF aims to optimize the prompt for a black-box LLM using only human preference feedback, is studied.
- The effects of a mid-task break on the brain connectome in healthy participants: A resting-state functional MRI study46
Understanding of how the resting brain reorganizes following the accumulation of prolonged task is augmented, dissociable processes between the neural mechanisms of fatigue and recovery are suggested, and some of the first quantitative insights into the cognitive neuroscience of work and rest are provided.
- Sample-Then-Optimize Batch Neural Thompson Sampling34
Two algorithms based on the Thompson sampling (TS) policy named Sample-Then-Optimize Batch Neural TS (STO-BNTS) and STO -BNTS-Linear are introduced and derive regret upper bounds for their algorithms with batch evaluations, and use insights from batch BO and NTK to show that they are asymptotically no-regret under certain conditions.
- Use Your INSTINCT: INSTruction optimization for LLMs usIng Neural bandits Coupled with Transformers28
A neural bandit algorithm is adopted which replaces the GP in BO by an NN surrogate to optimize instructions for black-box LLMs and is used to propose the INSTINCT algorithm, which consistently outperforms the existing methods in different tasks.
- Stress-induced changes in modular organizations of human brain functional networks28
It is demonstrated that acute stress prompts large-scale brain-wide reconfiguration involving multiple functional modules, corresponding to several well-known functional systems underpinning cognitive control, self-referential mental processing, visual, salience processing, sensory and motor functions.
- R2-B2: Recursive Reasoning-Based Bayesian Optimization for No-Regret Learning in Games28
It is proved that by reasoning at level 2 or more and at one level higher than the other agents, the R2-B2 agent can achieve faster asymptotic convergence to no regret than that without utilizing recursive reasoning.
- Temporal efficiency evaluation and small-worldness characterization in temporal networks28
A novel generalized analytical framework for temporal networks is introduced, which enables robust evaluation of the efficiency of temporal information exchange using two new network metrics and quantitative inspection of the temporal small-worldness, and can provide further insight into dynamic changes in the network topology of various real-world systems.
- Quantum Bayesian Optimization22
This work introduces the quantum-Gaussian process-upper confidence bound (Q-GP-UCB) algorithm, which is the first BO algorithm able to achieve a regret upper bound of O(polylog T), which is significantly smaller than its regret lower bound of Omega(sqrt(T)) in the classical setting.
- Source Attribution for Large Language Model-Generated Data21
This paper shows that both problems can be solved by watermarking, i.e., by enabling an LLM to generate synthetic texts with embedded watermarks that contain information about their source(s) and achieves effective source attribution and data provenance.
- Altered intra- and inter-hemispheric functional dysconnectivity in schizophrenia21
The source of the functional dysconnection in schizophrenia is examined, providing new insights into the nature of dysconnectivity in schizophrenia, and the dissociable processes between the preserved intra-hemispheric network topology and altered inter-hemisphere functional connectivity are highlighted.
- Bayesian Optimization under Stochastic Delayed Feedback20
This paper proposes algorithms with sub-linear regret guarantees that efficiently address the dilemma of selecting new function queries while waiting for randomly delayed feedback.
- Real-Time Aligned Reward Model beyond Semantics19
R2M (Real-Time Aligned Reward Model), a novel lightweight RLHF framework that goes beyond vanilla reward models that solely depend on the semantic representations of a pretrained LLM and leverages the evolving hidden states of the policy to align with the real-time distribution shift of the policy during the RL process.
- Adaptive Batch-Wise Sample Scheduling for Direct Preference Optimization15
SamS is proposed, an efficient and effective algorithm that adaptively selects samples in each training batch based on the LLM's learning feedback to maximize the potential generalization performance, without modifying the core DPO algorithm.
- 15
- Refining Adaptive Zeroth-Order Optimization at Ease14
This paper first shows the untapped variance reduction effect of first moment estimate on ZO gradient estimation, which improves the accuracy and stability of ZO updates, and refined the second moment estimate based on these variance-reduced gradient estimates to better capture the geometry of the optimization landscape, enabling a more effective scaling of ZO updates.
- Federated Neural Bandits14
The federated neural-upper confidence bound (FN-UCB) algorithm is introduced, which adopts a weighted combination of two UCBs: UCB a allows every agent to additionally use the observations from the other agents to accelerate exploration (without sharing raw observations); UCB b uses an NN with aggregated parameters for reward prediction in a similar way as federated averaging for supervised learning.
- On Provably Robust Meta-Bayesian Optimization13
It is proved that both meta-BO algorithms are asymptotically no-regret even when some or all previous tasks are dissimilar to the current task, and it is shown that RM-GP-UCB enjoys a better theoretical robustness than RM- GP-TS.
- 13
- ActiveDPO: Active Direct Preference Optimization for Sample-Efficient Alignment12
This work proposes an algorithm, ActiveDPO, that uses a theoretically grounded data selection criterion for non-linear reward functions while directly leveraging the LLM itself to parameterize the reward model used for active data selection, thereby leading to more effective and efficient data collection.
- Adjusted Expected Improvement for Cumulative Regret Minimization in Noisy Bayesian Optimization12
A high-probability regret upper bound of EIC is established based on the maximum information gain which is tighter than the bound of existing EI-based algorithms and comparable to the regret bound of other popular BO algorithms such as Thompson sampling and upper confidence bound.
- 12
- Robustifying and Boosting Training-Free Neural Architecture Search11
The robustifying and boosting training-free NAS (RoBoT) algorithm is proposed which employs the optimized combination of existing training-free metrics explored from Bayesian optimization to develop a robust and consistently better-performing metric on diverse tasks, and applies greedy search on the newly developed metric to bridge the gap.
- Federated Zeroth-Order Optimization using Trajectory-Informed Surrogate Gradients11
The FZooS achieves theoretical improvements over the existing approaches, which is supported by real-world experiments such as federated black-box adversarial attack and federated non-differentiable metric optimization.
- Batch Bayesian Optimization for Replicable Experimental Design10
The Batch Thompson Sampling for Replicable Experimental Design (BTS-RED) framework is proposed, which encompasses three algorithms that choose the number of replications adaptively rather than deterministically such that an input with a larger noise variance is replicated more times.
- 9
- Training-Free Neural Active Learning with Initialization-Robustness Guarantees8
This work empirically demonstrate that the expected variance with Gaussian processes (EV-GP) criterion is highly correlated with both initialization robustness and generalization performance, and show that it consistently outperforms baseline methods in terms of both desiderata, especially in situations with limited initial data or large batch sizes.
- Recursive reasoning-based training-time adversarial machine learning8
It is shown how an R2T2 attacker (defender) can utilize the proposed nested projected gradient descent-based method to approximate the optimal attack (defense) strategy at an arbitrary level of reasoning.
- 8
- 8
- MASPOB: Bandit-Based Prompt Optimization for Multi-Agent Systems with Graph Neural Networks7
MASPOB (Multi-Agent System Prompt Optimization via Bandits), a novel sample-efficient framework based on bandits that achieves state-of-the-art performance, consistently outperforming existing baselines.
- Meta-Prompt Optimization for LLM-Based Sequential Decision Making7
The EXPonential-weight algorithm for prompt Optimization} (EXPO) is proposed to automatically optimize the task description and meta-instruction in the meta-prompt for LLM-based agents and is extended to additionally optimize the exemplars (i.e., history of interactions) in the meta-prompt to further enhance the performance, hence introducing the EXPO-ES algorithm.
- T-POP: Test-Time Personalization with Online Preference Feedback7
This work proposes T-POP (Test-Time Personalization with Online Preference Feedback), a novel algorithm that synergistically combines test-time alignment with dueling bandits that achieves rapid and data-efficient personalization, significantly outperforming existing baselines and showing consistent improvement with more user interactions.
- Dual-Robust Cross-Domain Offline Reinforcement Learning Against Dynamics Shifts7
A novel robust cross-domain Bellman (RCB) operator is introduced, which enhances test-time robustness against dynamics perturbations while staying conservative to the out-of-distribution dynamics transitions, thus guaranteeing the train-time robustness.
- 7
- 7
- 7
- DemoPSD: Disagreement-Modulated Policy Self-Distillation6
DemoPSD is introduced, a novel framework that resolves problems through the idea of *selective adoption of teacher guidance*.
- Exploiting Correlated Auxiliary Feedback in Parameterized Bandits6
This paper develops a method that exploits auxiliary feedback to build a reward estimator with tight confidence bounds, leading to a smaller regret and describes the regret reduction in terms of the correlation coefficient between reward and its auxiliary feedback.
- EduAgentQG: Multi-Agent Personalized Mathematics Question Generation with Explicit Diversity and Objective-Aware Evaluation5
E EduAgentQG, a multi-agent collaborative framework for generating high-quality and diverse personalized questions that outperforms existing single-agent and multi-agent methods in terms of question diversity, goal consistency, and overall quality.
- Data-Centric AI in the Age of Large Language Models5
A data-centric viewpoint of AI research, focusing on large language models (LLMs), is proposed, covering data-centric benchmarks and data curation, data attribution, knowledge transfer, knowledge transfer, and inference contextualization.
- Active Human Feedback Collection via Neural Contextual Dueling Bandits4
It is theoretically show that when preference feedback follows the Bradley-Terry-Luce model, the worst sub-optimality gap of the policy learned by Neural-ADB decreases at a sub-linear rate as the preference dataset increases.
- 4
- Cognitive Workload Discrimination in Flight Simulation Task Using a Generalized Measure of Association4
A novel, parameter-free method of evaluating the cortical connectivity, named Generalized Measure of Association (GMA), is adopted to assess and discriminate the mental workload of pilots using the Multi-Attribute Task Battery (MATB) flight simulation platform.
- Harness-MU: A Safe, Governed, and Effective Harness for Multi-User LLM Agents3
Harness-MU is presented, the first model-agnostic, zero-tuning infrastructure framework for multi-user LLM agents, establishing that systematic infrastructure is essential for solving LLM multi-principal governance challenges.
- FedPOB: Sample-Efficient Federated Prompt Optimization via Bandits3
This work proposes the Federated Prompt Optimization via Bandits (FedPOB) algorithm, a federated variant of the Linear UCB algorithm, where agents collaborate by sharing model parameters instead of raw data, and extends the approach to the practical setting of comparative user feedback by introducing FedPOB with Preference Feedback (FedPOB-Pref), an efficient algorithm based on federated dueling bandits.
- Online Clustering of Dueling Bandits3
Two novel algorithms are proposed which model the user reward functions as linear functions of the context vectors and uses a neural network to model complex, non-linear user reward functions to enable collaborative decision-making based on preference feedback.
- 3
- Learning to Reason with Insight for Informal Theorem Proving2
This work proposes a unified training framework designed to cultivate this essential reasoning skill and enable LLMs to perform insightful reasoning, and demonstrates that teaching models to identify and apply core techniques can substantially improve their mathematical reasoning.
- 2
- 2
- 2
- 2
- Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation1
ROSA2 is proposed, a framework that reformulates interaction as a joint optimization problem over the heterogeneous space of Words and Weights, and it is proved that this co-adaptation strictly reduces the required parameter shift for convergence.
- CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models1
This work proposes the concept of Student-Tailored Personalized Safety and construct CASTLE based on educational theories, and designs three evaluation metrics: Risk Sensitivity, measuring the model ability to detect risks; Emotional Empathy, evaluating the model capacity to recognize student states; and Student Alignment, assessing the match between model responses and student attributes.
- UniScale: Adaptive Unified Inference Scaling via Online Joint Optimization of Model Routing and Test-Time Scaling1
UIS is introduced, an online framework that models adaptive UIS as a contextual multi-armed bandit problem and learns inference policies via LinUCB and effectively exploits the synergy in the UIS space to deliver a fine-grained and consistently better quality-cost trade-off across diverse, dynamic inference scenarios.
- 1
- 1
- 1
- –
- Federated Linear Dueling Bandits–
Through rigorous theoretical analysis, it is proved that FLDB-OGD enjoys a sub-linear upper bound on its cumulative regret and a theoretical trade-off between regret and communication complexity is demonstrated.
- –
- –
- –
- –
- DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization–
DRIFT (Decoupled Rollouts and Importance-Weighted Fine-Tuning) is proposed, a framework that operationalizes the theoretical insight that the KL-regularized RL objective is equivalent to importance-weighted supervised learning.
- –
- Why Zeroth-Order Adaptation May Forget Less: A Randomized Shaping Theory–
RISE applies the calibrated ZO shape to exact FO gradients inside parameter blocks as an algorithmic transfer that is a stability--plasticity tradeoff, and improves mean forgetting when the FO direction has above-average retention curvature.
- MetaForge: A Self-Evolving Multimodal Agent that Retrieves, Adapts, and Forges Tools On Demand–
MetaForge is proposed, a multimodal agent framework that learns when to invoke tools and how to evolve its toolset on demand, validating a paradigm shift from static tool inventories to on-demand self-evolution.
- CES: Combinatorial Experts Selection via Contextual Linear Bandits–
This work proposes an offline-to-online combinatorial experts selection (CES) framework that leverages offline evaluation to warm-start model capability estimation and employs online learning to adapt to capability shifts and correct offline inaccuracies.
- Linear and Neural Dueling Bandits with Delayed Feedback–
This work formalizes the problem of Contextual Dueling Bandits with Stochastic Delayed Feedback and proposes two novel algorithms: Linear (LDB-DF) and Neural (NDB-DF) Dueling Bandits with Delayed Feedback, a novel estimator that integrates an Inverse Probability Weighting mechanism directly into the loss function, ensuring unbiased correction for delayed or missing feedback.
- –
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.