Recent progress in o1-like models has significantly enhanced the reasoning abilities of Large Language Models (LLMs), empowering them to tackle increasingly complex tasks through reflection capabilities, such as making assumptions, backtracking, and self-refinement. However, effectively evaluating such reflection capabilities remains challenging due to the lack of appropriate benchmarks. To bridge this gap, we introduce LR$^2$Bench, a novel benchmark designed to evaluate the Long-chain Reflective Reasoning capabilities of LLMs. LR$^2$Bench comprises 850 samples across six Constraint Satisfaction Problems (CSPs) where reflective reasoning is crucial for deriving solutions that meet all given constraints. Each type of task focuses on distinct constraint patterns, such as knowledge-based, logical, and spatial constraints, providing a comprehensive evaluation of diverse problem-solving scenarios. We conduct extensive evaluation on both conventional models and o1-like models. Our experimental results reveal that even the most advanced reasoning-specific models, such as DeepSeek-R1 and OpenAI o1-preview, struggle with tasks in LR$^2$Bench, achieving an average Exact Match score of only 20.0% and 23.6%, respectively. These findings underscore the significant room for improvement in the reflective reasoning capabilities of current LLMs. The leaderboard of our benchmark is available at https://huggingface.co/spaces/UltraRonin/LR2Bench

Constrained Online Convex Optimization (COCO) can be seen as a generalization of the standard Online Convex Optimization (OCO) framework. At each round, a cost function and constraint function are revealed after a learner chooses an action. The goal is to minimize both the regret and cumulative constraint violation (CCV) against an adaptive adversary. We show for the first time that is possible to obtain the optimal $O(\sqrt{T})$ bound on both regret and CCV, improving the best known bounds of $O \left( \sqrt{T} \right)$ and $\~{O} \left( \sqrt{T} \right)$ for the regret and CCV, respectively.

In this paper, we study bi-criteria optimization for combinatorial multi-armed bandits (CMAB) with bandit feedback. We propose a general framework that transforms discrete bi-criteria offline approximation algorithms into online algorithms with sublinear regret and cumulative constraint violation (CCV) guarantees. Our framework requires the offline algorithm to provide an $(\alpha, \beta)$-bi-criteria approximation ratio with $\delta$-resilience and utilize $\texttt{N}$ oracle calls to evaluate the objective and constraint functions. We prove that the proposed framework achieves sub-linear regret and CCV, with both bounds scaling as ${O}\left(\delta^{2/3} \texttt{N}^{1/3}T^{2/3}\log^{1/3}(T)\right)$. Crucially, the framework treats the offline algorithm with $\delta$-resilience as a black box, enabling flexible integration of existing approximation algorithms into the CMAB setting. To demonstrate its versatility, we apply our framework to several combinatorial problems, including submodular cover, submodular cost covering, and fair submodular maximization. These applications highlight the framework's broad utility in adapting offline guarantees to online bi-criteria optimization under bandit feedback.

FastMap was first introduced in the Data Mining community for generating Euclidean embeddings of complex objects. In this dissertation, we first present FastMap to generate Euclidean embeddings of graphs in near-linear time: The pairwise Euclidean distances approximate a desired graph-based distance function on the vertices. We then apply the graph version of FastMap to efficiently solve various graph-theoretic problems of significant interest in AI: including facility location, top-K centrality computations, community detection and block modeling, and graph convex hull computations. We also present a novel learning framework, called FastMapSVM, by combining FastMap and Support Vector Machines. We then apply FastMapSVM to predict the satisfiability of Constraint Satisfaction Problems and to classify seismograms in Earthquake Science.

The application of learning based methods to vehicle routing problems has emerged as a pivotal area of research in combinatorial optimization. These problems are characterized by vast solution spaces and intricate constraints, making traditional approaches such as exact mathematical models or heuristic methods prone to high computational overhead or reliant on the design of complex heuristic operators to achieve optimal or near optimal solutions. Meanwhile, although some recent learning-based methods can produce good performance for VRP with straightforward constraint scenarios, they often fail to effectively handle hard constraints that are common in practice. This study introduces a novel end-to-end framework that combines constraint-oriented hypergraphs with reinforcement learning to address vehicle routing problems. A central innovation of this work is the development of a constraint-oriented dynamic hyperedge reconstruction strategy within an encoder, which significantly enhances hypergraph representation learning. Additionally, the decoder leverages a double-pointer attention mechanism to iteratively generate solutions. The proposed model is trained by incorporating asynchronous parameter updates informed by hypergraph constraints and optimizing a dual loss function comprising constraint loss and policy gradient loss. The experiment results on benchmark datasets demonstrate that the proposed approach not only eliminates the need for sophisticated heuristic operators but also achieves substantial improvements in solution quality.

Constraints are critical in text generation as LLM outputs are often unreliable when it comes to ensuring generated outputs adhere to user defined instruction or general safety guidelines. To address this gap, we present Constrained Discrete Diffusion (CDD), a novel method for enforcing constraints on natural language by integrating discrete diffusion models with differentiable optimization. Unlike conventional text generators, which often rely on post-hoc filtering or model retraining for controllable generation, we propose imposing constraints directly into the discrete diffusion sampling process. We illustrate how this technique can be applied to satisfy a variety of natural language constraints, including (i) toxicity mitigation by preventing harmful content from emerging, (ii) character and sequence level lexical constraints, and (iii) novel molecule sequence generation with specific property adherence. Experimental results show that our constraint-aware procedure achieves high fidelity in meeting these requirements while preserving fluency and semantic coherence, outperforming auto-regressive and existing discrete diffusion approaches.

Algorithmic Recourse is a way for users to modify their attributes to align with a model's expectations, thereby improving their outcomes after receiving unfavorable decisions. In real-world scenarios, users often need to strategically adjust their attributes to compete for limited resources. However, such strategic behavior induces users to "game" algorithms, causing model collapse due to distribution shifts. These shifts arise from user competition, resource constraints, and adaptive user responses. While prior research on Algorithmic Recourse has explored its effects on both systems and users, the impact of resource constraints and competition over time remains underexplored. In this work, we develop a general framework to model user strategic behaviors and their interactions with decision-making systems under resource constraints and competitive dynamics. Through theoretical analysis and empirical evaluation, we identify three key phenomena that arise consistently in both synthetic and real-world datasets: escalating decision boundaries, non-robust model predictions, and inequitable recourse actions. Finally, we discuss the broader social implications of these findings and present two algorithmic strategies aimed at mitigating these challenges.

This study presents a methodology to safely manipulate branches to aid various agricultural tasks. Humans in a real agricultural environment often manipulate branches to perform agricultural tasks effectively, but current agricultural robots lack this capability. This proposed strategy to manipulate branches can aid in different precision agriculture tasks, such as fruit picking in dense foliage, pollinating flowers under occlusion, and moving overhanging vines and branches for navigation. The proposed method modifies RRT* to plan a path that satisfies the branch geometric constraints and obeys branch deformable characteristics. Re-planning is done to obtain a path that helps the robot exert force within a desired range so that branches are not damaged during manipulation. Experimentally, this method achieved a success rate of 78% across 50 trials, successfully moving a branch from different starting points to a target region.

We propose a novel, multi-layered planning approach for computing paths that satisfy both kinodynamic and spatiotemporal constraints. Our three-part framework first establishes potential sequences to meet spatial constraints, using them to calculate a geometric lead path. This path then guides an asymptotically optimal sampling-based kinodynamic planner, which minimizes an STL-robustness cost to jointly satisfy spatiotemporal and kinodynamic constraints. In our experiments, we test our method with a velocity-controlled Ackerman-car model and demonstrate significant efficiency gains compared to prior art. Additionally, our method is able to generate complex path maneuvers, such as crossovers, something that previous methods had not demonstrated.

Guaranteeing constraint satisfaction is challenging in imitation learning (IL), particularly in tasks that require operating near a system's handling limits. Traditional IL methods often struggle to enforce constraints, leading to suboptimal performance in high-precision tasks. In this paper, we present a simple approach to incorporating safety into the IL objective. Through simulations, we empirically validate our approach on an autonomous racing task with both full-state and image feedback, demonstrating improved constraint satisfaction and greater consistency in task performance compared to a baseline method.