Searcharxiv⌕ Search

arXiv subjects

Somayeh Sojoudi

Publications and source records attributed to Somayeh Sojoudi.

At least 19 recordsLinked to original sources

Distilling Safety Filters for Filter-free Multi-agent RL

Reinforcement learning (RL) can teach robot teams to navigate shared spaces, but trial and error causes collisions that real robots cannot afford. Control barrier function (CBF) safety filters prevent these collisions by correcting unsafe actions, yet in a team the filter becomes a centralized optimization whose number of constraints grows quadratically with the team size. We propose an approach that builds the filter's safety into the RL policy and examine how such a system behaves during filter-free deployment. However, the filter that keeps training safe also hides what the policy must learn: it masks the cost of unsafe actions, which actor-critic methods then overvalue, and it keeps near-collision states out of the training data. Our approach therefore learns only from executed safe actions with implicit Q-learning and distills the filter into the policy at synthetic near-collision states. We prove that filter distillation loss bounds the policy's constraint violation and hence how far agents can cross their safety margin, so errors never accumulate. When distillation states cover deployed states, the loss also bounds the collision probability at every step. Without a filter and acting only on local observations, our policies have no collisions in 1,800 test episodes with multiple agents. A policy trained with three agents transfers zero-shot to teams of twenty at constant per-agent computation and stays collision-free with up to eight agents. We test our policies on Crazyflie quadrotors with up to five flying at the same time.

cs.RO↗

Transformers Provably Learn Graph Search: Training Dynamics and the Exponential Power of Depth

Large Language Models (LLMs) have achieved remarkable progress on complex reasoning tasks, yet the mechanisms by which they acquire reasoning capabilities and the efficiency of their reasoning remain poorly understood. In this paper, we investigate the path-finding problem over directed graphs, a symbolic abstraction of multi-step reasoning. We first characterize the architectural requirements for this task, proving that one-layer transformers require $Ω(N^2)$ size on dense graphs to implement the key DFS child-selection primitive. While a two-layer transformer implements full DFS with $O(N\log(N))$ size. Thus, a second layer is necessary and sufficient for near-linear DFS reasoning. We then extend this expressivity analysis to deeper models: for every $k\geq3$, a $k$-layer transformer implements a bounded-reachability DFS policy with look-ahead horizon $1+2^{k-3}$, establishing an exponential increase in reachability horizon with depth. Finally, going beyond expressivity, given appropriately curated training data, we show that two-layer transformers can provably learn to execute DFS via gradient flow and generalize to unseen graphs. Together, our results provide a mechanistic account of how transformers implement graph search, how depth expands bounded look-ahead, and how such reasoning algorithms can emerge through training.

cs.AI↗

Strategic Arbitrage in Electricity Markets: A Multi-Agent Policy Optimization Framework

Grid-scale storage and load aggregators are growing quickly in wholesale electricity markets. If highly consolidated, these new participants may distort the competitive market equilibrium through strategic withholding behavior. Traditional market power mitigation techniques are ill-suited to arbitrage-based participants. We provide a multi-agent policy optimization model to study game-theoretic equilibria for real-time markets with consolidated storage and/or aggregators. The model is designed to facilitate natural first-order equilibrium-seeking update rules. We analyze the theoretical properties of different supply functions and their implications on the convergence of the gradient dynamics. Finally, we train equilibrium policies for a modified version of the IEEE RTS-GMLC benchmark across different levels of storage fleet consolidation, to demonstrate the potential of the framework as a tool to characterize and quantify strategic arbitrage.

eess.SY↗

Intent Demonstration in General-Sum Dynamic Games via Iterative Linear-Quadratic Approximations

Autonomous agents should coordinate effectively without prior knowledge of others' intents. While prior work has focused on intent inference, we address the inverse problem: how agents can strategically demonstrate their intents within general-sum dynamic games. We model this problem and propose an algorithm that balances intent demonstration with task performance. To handle nonlinear dynamic games with continuous state-action spaces, our method leverages iterative linear-quadratic game approximations and provides efficient intent-teaching guarantees: the uncertain agent's belief can be driven rapidly to the ground truth, while the demonstrating agent avoids expending effort on unnecessary belief alignment when it does not improve task performance. Theoretical analysis and hardware experiments confirm that our approach enables the demonstrating agent to reconcile task execution with belief alignment and strategically manage the information asymmetry among agents, even as its intent evolves during deployment.

eess.SY↗

A Deficiency-Based Approach for Unbalanced Three-Phase Distribution System Expansion Planning

Distribution planners typically characterize grid needs through power flow studies and fit upgrades to those needs by hand. More principled one-shot optimization methods either do not scale or misrepresent unbalanced three-phase networks. This paper proposes an iterative planning process that alternates between two problem classes with mature off-the-shelf solvers. An optimal power flow (OPF) problem quantifies the deficiency of a candidate portfolio by optimizing the dispatch of controllable devices and measuring thermal and voltage violations on a common scale, with an optimal value of zero exactly at adequate portfolios. The OPF multipliers give a first-order approximation of the relief each investment candidate would provide. We bound the error of this approximation and show that structural properties of the deficiency guard it against overestimation. Building on this property, a heuristic cutting-plane search uses a mixed-integer linear program to propose each next portfolio from these approximations. On six unbalanced three-phase feeders from the SMART-DS dataset, the search outperforms a linearized optimization-based benchmark, and the results agree across five variants of the deficiency.

eess.SY↗

Verifier-Grounded Agentic RL for Provably Correct Code

Reinforcement learning for code usually rewards passing tests, which a model can do while remaining wrong on the inputs the tests do not cover. A program verifier instead decides whether a program satisfies a formal specification for all inputs. Language models are almost always paired with verifiers at inference time only, and turning a verifier into a reward raises two difficulties. It reports only acceptance or rejection, which gives no gradient when every attempt fails, and a policy can cheat it, most simply by telling it to assume what the proof was supposed to establish. We build an agentic environment and reward that address both. The reward reads the verifier's diagnostics rather than its verdict, scoring a candidate by the weighted fraction of proof obligations it discharges, extended with hidden probes from a reference proof's annotations. An integrity audit outranks the verifier, rejecting any candidate that alters the specification or assumes what it was meant to prove. The model works as an agent against a live verifier over many turns, with a separately trained proof specialist as a frozen collaborator. Across Qwen3 models from 1.7B to 14B and a 35B Qwen3.6 mixture-of-experts model, held-out verification on code generation rises from 66.1% to 93.5% on DafnyBench and from 56.3% to 100% on SWE-Proof. This reward leads every other training signal we compare, including a binary pass-or-fail reward trained under otherwise identical settings, with the widest margin on code generation. When the agent must also deliver the repository patch, graded by each project's withheld tests, the fraction of held-out issues it resolves rises from 59.6% to 72.5%. Evaluated unchanged on three verification suites withheld from training, the checkpoints improve on all of them, including one whose language and verifier are absent from training.

cs.LG↗

Multi-Agent Guided Policy Search for Non-Cooperative Dynamic Games

Multi-agent reinforcement learning (MARL) optimizes strategic interactions in non-cooperative dynamic games, where agents have misaligned objectives. However, data-driven methods such as multi-agent policy gradients (MA-PG) often suffer from instability and limit-cycle behaviors. Prior stabilization techniques typically rely on entropy-based exploration, which slows learning and increases variance. We propose a model-based approach that incorporates approximate priors into the reward function as regularization. In linear quadratic (LQ) games, we prove that such priors stabilize policy gradients and guarantee local exponential convergence to an approximate Nash equilibrium. We then extend this idea to infinite-horizon nonlinear games by introducing Multi-agent Guided Policy Search (MA-GPS), which constructs short-horizon local LQ approximations from trajectories of current policies to guide training. Experiments on nonlinear vehicle platooning and a six-player strategic basketball formation show that MA-GPS achieves faster convergence and more stable learning than existing MARL methods.

eess.SY↗

Scenario-Game ADMM: A Parallelized Scenario-Based Solver for Stochastic Noncooperative Games

Decision-making in multi-player games can be extremely challenging, particularly under uncertainty. In this work, we propose a new sample-based approximation to a class of stochastic, general-sum, pure Nash games, where each player has an expected-value objective and a set of chance constraints. This new approximation scheme inherits the accuracy of objective approximation from the established sample average approximation (SAA) method and enjoys a feasibility guarantee derived from the scenario optimization literature. We characterize the sample complexity of this new game-theoretic approximation scheme, and observe that high accuracy usually requires a large number of samples, which results in a large number of sampled constraints. To accommodate this, we decompose the approximated game into a set of smaller games with few constraints for each sampled scenario, and propose a decentralized, consensus-based ADMM algorithm to efficiently compute a generalized Nash equilibrium (GNE) of the approximated game. We prove the convergence of our algorithm to a GNE and empirically demonstrate superior performance relative to a recent baseline algorithm based on ADMM and interior point method.

eess.SY↗

Recursive Self-Improvement through Multi-Agent Self-Supervision

Recursive self-improvement (RSI) of a model on non-verifiable tasks, such as open-ended research, faces a supervision bottleneck when its outputs exceed what even human experts can reliably assess, leaving the model itself (optimizee) as the best available optimizer and evaluator. However, a single model instance struggles to critique and improve its own complex reasoning under this homogeneous loop. To address this, we propose Multi-Agent Self-Supervision (MASS), an RSI method that alternates between evolutionary workflow optimization and supervised fine-tuning on self-generated trajectories. Guided by early findings that multi-agent topologies excel at complex reasoning, MASS prompts a single base model to iteratively propose, execute, and self-evaluate multi-agent workflows. Through an evolutionary search constrained by structural guardrails, the model optimizes these computational-graph-like orchestrations, discovering the most effective distinct roles and information routing for a given task. Over two MASS cycles with Qwen3.6-27B, the model achieves 1.2-1.6x higher performance per output tokens on four open-ended public benchmarks. Because the improved model subsequently acts as a better optimizer and evaluator, this alternating framework enables a continuous, recursive bootstrapping of the model's capabilities. Moreover, multi-agent traces are also more training-efficient: a student trained on them outperforms a single-agent student trained on 1.4x more training tokens. These findings suggest that jointly learning orchestration and bounded subagent execution from multi-agent trajectories can provide an effective signal for RSI.

cs.AI↗

How RL Reshapes LLM Reasoning: Transferability, Coverage, and Scaling Laws

Recent studies on reinforcement learning (RL) report seemingly conflicting evidence about large language model (LLM) reasoning. Training on mathematics can improve performance in other domains, yet gains in Pass@1 can coincide with lower Pass@$N$ than the base model. This raises a fundamental question: does RL expand an LLM's reasoning boundary, or merely reweight its existing reasoning space? We revisit these phenomena across Qwen and Gemma model families, showing both cross-domain gains and forgetting, while coverage at large sampling budgets increases on some tasks and decreases on others. Detailed analysis of solution traces before and after RL indicates a shift in the reasoning strategies the model employs, motivating a two-stage autoregressive policy model that separates \emph{strategy selection} from problem-specific execution. Within this framework, we prove how RL's implicit bias reshapes strategy preferences, allowing gains on some tasks while suppressing strategies required by others. This mechanism can also broaden or narrow coverage at a given sampling budget even without expanding strategy support. We further provide theoretical justifications for log-sigmoid and log-linear scaling laws in RL compute, and evaluate their predictive power. Together, these results connect changes in strategy selection to cross-domain transfer, reasoning coverage, and compute scaling.

cs.LG↗

SWE-Proof: Can Language Models Resolve Real-World Issues with Machine-Checked Proofs?

Ensuring the correctness of LLM-generated code is a core challenge for modern software engineering. Benchmarks for agentic code generation check correctness with held-out test suites, which are inherently incomplete and increasingly susceptible to memorization. Formal verification avoids both problems, but existing work covers only standalone tasks whose specifications are given as input, not real issues, which touch large repositories and state intent in vague natural language. We present Benchproofer, a pipeline that turns a coding task with a known correct patch into a formally verified one: it writes a specification for the new code, summarizes the existing functions that code calls with axioms, and admits an instance only after mechanical and adversarial gates agree. Applying it to SWE-bench Verified yields SWE-Proof, 500 real issues whose correctness is formally verified rather than tested, and it extends to SWE-bench Pro. Evaluating Claude Opus 4.8, we find that verification catches what tests miss: a quarter of test-passing patches admit counterexamples, which a structured natural-language specification does not fix, while a correct formal one lifts resolution from 85% to 95%. Writing that specification is the hard part: an agent that must write its own gains nothing over an unaided baseline, and only 56% of those specifications pass our audit. The usual failure is faithfulness, a specification that constrains part of the required behavior and leaves the rest free. Specification quality still tracks the outcome, failing on 92% of unresolved instances against 51% of resolved ones, making faithful specification synthesis a concrete open problem.

cs.LG↗

Breaking the Reversal Curse in Autoregressive Language Models via Identity Bridge

Autoregressive large language models (LLMs) have achieved remarkable success in many complex tasks, yet they can still fail in very simple logical reasoning such as the "reversal curse" -- when trained on forward knowledge data of the form "$A \rightarrow B$" (e.g., Alice's husband is Bob), the model is unable to deduce the reversal knowledge "$B \leftarrow A$" (e.g., Bob's wife is Alice) during test. Extensive prior research suggests that this failure is an inherent, fundamental limit of autoregressive causal LLMs, indicating that these models tend to memorize factual-level knowledge rather than capture higher-level rules. In this paper, we challenge this view by showing that this seemingly fundamental limit can be mitigated by slightly tweaking the training data with a simple regularization data recipe called the Identity Bridge of the form "$A \to A$" (e.g., The name of Alice is Alice). Theoretically, we prove that under this recipe, even a one-layer transformer can break the reversal curse by analyzing the implicit bias of gradient descent. Empirically, we show that a 1B pretrained language model finetuned with the proposed data recipe achieves a 50% success rate on reversal tasks, in stark contrast to a near-zero success rate when trained solely on forward-knowledge data. Our work provides a novel theoretical foundation for the reversal curse and offers a principled, low-cost path to encouraging LLMs to learn higher-level rules from data.

cs.AI↗

Transformers Provably Learn to Internalize Chain-of-Thought

Chain-of-Thought (CoT) prompting substantially improves the sample efficiency of transformers, reducing the complexity of tasks like parity learning from exponential to polynomial in the input length. However, generating explicit reasoning steps at inference is computationally expensive. Implicit Chain-of-Thought (ICoT) has emerged as a promising empirical remedy that trains models to internalize intermediate steps within their hidden states, but its theoretical foundations remain poorly understood. We give the first theoretical analysis of ICoT, proving that an $L$-layer transformer trained under our proposed Log-ICoT curriculum learns $k$-parity with $\mathsf{poly}(n)$ samples and $L = \log_2 k$ training stages. This matches the sample efficiency of explicit CoT while eliminating its inference overhead, and extends prior one-layer parity guarantees to multi-layer architectures. Compared to standard ICoT, which removes thinking tokens one at a time, Log-ICoT removes them in geometric chunks, reducing the number of stages from linear in $k$ to logarithmic. Experiments on multi-layer transformers confirm the theory and visualize how reasoning is progressively absorbed into deeper layers.

cs.LG↗

Multi-Objective Learning for Diffusion Models: A Statistical Theory under Semi-Supervised Learning

Diffusion models are increasingly used as powerful conditional generators, yet real deployments often involve multiple target distributions arising from different tasks, e.g., diverse prompt domains in text-to-image generation, or multiple environments in robotics with diffusion policies. This naturally leads to a multi-objective learning (MOL) problem. A key challenge is that achieving good Pareto trade-offs can require a generalist model class with substantially larger capacity than what suffices for solving any individual task, thereby increasing statistical cost since sample complexity typically scales with the model complexity. To reconcile this, we develop a principled MOL framework for diffusion models with limited data: a semi-supervised regime where paired (labeled) samples are scarce, but (unlabeled) condition data are abundant. We propose a two-stage training procedure that first fits lightweight specialist models from limited paired data, and then distills them into a generalist model by generating pseudo-samples. We establish generalization bounds showing that the required number of paired samples only depends on the complexity of the specialist model classes. We further extend the theory to diffusion policies for sequential decision making to account for distribution shift in on-policy rollouts. Extensive experiments on robotic control and image restoration tasks are conducted to verify our theoretical results.

cs.LG↗

Do Sparse Autoencoders Identify Reasoning Features in Language Models?

We study how reliably sparse autoencoders (SAEs) support claims about reasoning-related internal features in large language models. We first give a stylized analysis showing that sparsity-regularized decoding can preferentially retain stable low-dimensional correlates while suppressing high-dimensional within-behavior variation, motivating the possibility that contrastively selected "reasoning" features may concentrate on cue-like structure when such cues are coupled with reasoning traces. Building on this perspective, we propose a falsification-based evaluation framework that combines causal token injection with LLM-guided counterexample construction. Across 22 configurations spanning multiple model families, layers, and reasoning datasets, we find that many contrastively selected candidates are highly sensitive to token-level interventions, with 45%-90% activating after injecting only a few associated tokens into non-reasoning text. For the remaining context-dependent candidates, LLM-guided falsification produces targeted non-reasoning inputs that trigger activation and meaning-preserving paraphrases of top-activating reasoning traces that suppress it. A small steering study yields minimal changes on the evaluated benchmarks. Overall, our results suggest that, in the settings we study, sparse decompositions can favor low-dimensional correlates that co-occur with reasoning, underscoring the need for falsification when attributing high-level behaviors to individual SAE features. Code is available at https://github.com/GeorgeMLP/reasoning-probing.

cs.LG↗

ScribbleEdit: Synthetic Data for Image Editing with Scribbles and Text

Recent progress in generative models has significantly advanced image editing capabilities, yet precise and intuitive user control remains difficult. Specifically, users often struggle to communicate both exact spatial layouts and specific semantic details simultaneously. While natural language instructions effectively convey high-level semantics like texture and color, they lack spatial specificity. Conversely, freehand scribbles provide rough spatial boundaries but cannot express detailed visual attributes. Consequently, achieving precise control requires combining both modalities. However, existing models struggle to jointly interpret abstract scribbles alongside text due to a lack of specialized training data. In this work, we introduce ScribbleEdit, a large-scale synthetic dataset designed to bridge this gap by combining natural language instructions with freehand scribble inputs for more accurate, controllable edits. We construct this dataset through a synthetic pipeline that automatically generates source-target image pairs via inpainting, which are then paired with human-drawn scribbles and VLM-generated text instructions. Using ScribbleEdit, we evaluate and finetune both diffusion-based and autoregressive unified multimodal image editing models. Our experiments reveal that while off-the-shelf models struggle with abstract scribble inputs, finetuning on our synthetic dataset significantly improves their ability to generate spatially aligned and semantically consistent edits.

cs.CV↗

Reinforcement Learning via Value Gradient Flow

We study behavior-regularized reinforcement learning (RL), where regularization toward a reference distribution (the dataset in offline RL or the base model in LLM RL finetuning) is essential to prevent value over-optimization caused by erroneous out-of-distribution extrapolation. Existing methods either rely on reparameterized policy gradient, which are difficult to scale to large generative models, or on reject sampling, which can be overly conservative when attempting to move beyond the behavior support. In this paper, we propose Value Gradient Flow (VGF), a scalable new paradigm for behavior-regularized RL. VGF casts behavior-regularized RL as an optimal transport problem that maps the reference distribution to the value-induced optimal policy distribution. We solve this transport problem via discrete gradient flow, where value gradients guide particles initialized from the reference distribution. Our analysis shows that VGF imposes regularization implicitly by controlling the transport budget. VGF eliminates explicit policy parameterization while remaining expressive and flexible, this enables adaptive test-time scaling by adjusting the transport budget. Extensive experiments demonstrate that VGF significantly outperforms prior methods, achieving state-of-the-art results on offline RL benchmarks (D4RL, OGBench) and LLM RL tasks. Code and runs can be found at https://ryanxhr.github.io/vgf.

cs.LG↗

Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers

Large language models (LLMs) can acquire new knowledge through fine-tuning, but this process exhibits a puzzling duality: models can generalize remarkably from new facts, yet are also prone to hallucinating incorrect information. However, the reasons for this phenomenon remain poorly understood. In this work, we argue that both behaviors stem from a single mechanism known as out-of-context reasoning (OCR): the ability to deduce implications by associating concepts, even those without a causal link. Our experiments across five prominent LLMs confirm that OCR indeed drives both generalization and hallucination, depending on whether the associated concepts are causally related. To build a rigorous theoretical understanding of this phenomenon, we then formalize OCR as a synthetic factual recall task. We empirically show that a one-layer single-head attention-only transformer with factorized output and value matrices can learn to solve this task, while a model with combined weights cannot, highlighting the crucial role of matrix factorization. Our theoretical analysis shows that the OCR capability can be attributed to the implicit bias of gradient descent, which favors solutions that minimize the nuclear norm of the combined output-value matrix. This mathematical structure explains why the model learns to associate facts and implications with high sample efficiency, regardless of whether the correlation is causal or merely spurious. Ultimately, our work provides a theoretical foundation for understanding the OCR phenomenon, offering a new lens for analyzing and mitigating undesirable behaviors from knowledge injection.

cs.CL↗