Searcharxiv⌕ Search

arXiv subjects

Rohan Pandey

Publications and source records attributed to Rohan Pandey.

At least 19 recordsLinked to original sources

Wrong the Same Way: Cross-Prompt Sharing Both Suppresses and Amplifies Reward Hacking in RLVR

Two accounts of how reinforcement learning with verifiable rewards (RLVR) interacts with an imperfect verifier coexist: in one, reward hacking is inherited from the base model; in the other, RL amplifies it. We study both in a controlled environment small enough to compute the entire response distribution exactly. We vary the universality of a leak, or the fraction of prompts that accept the same wrong response, and the verifier's static false-positive rate. At its KL-regularized optimum, RLVR acts as a filter that reweights the accepted set without altering its composition, so the hack rate is inherited exactly. Taking that as a null model, a known decomposition writes the drift from it as a sum over prompts of covariances between reward and the empirical tangent kernel. The terms from other prompts vanish without parameter sharing. With sharing, every prompt that accepts a wrong answer pushes it up and every prompt that rejects it pushes it down through the baseline, in proportion to the model's probability of producing it there. A shared-parameter policy thus drives a leak that few prompts accept below the base rate and amplifies one that most prompts accept. At every convergent learning rate under Adam, SGD, and sampled GRPO, hacking ends higher at high universality than at low, but whether either end crosses the base rate depends on the step size. Independent per-prompt copies remove the dependence. The direction of this push depends on neither the verifier's error rate nor universality alone, but on how much of the model's kernel-weighted probability of a wrong answer falls on prompts that accept it, which frozen base model controls support. Verifier fixes should target the mistakes the model repeats across prompts: two patches that remove equally many errors differ by up to 38 points of true accuracy. We propose this alignment as a hypothesis for why earlier studies disagree.

cs.LG↗

MUSES: A Benchmark for Prospective Intellectual-Roots Retrieval

Scientific discovery depends on finding prior literature that shapes what comes next. Existing retrieval systems optimize for relevance and popularity, often favoring central papers over less familiar works that later prove generative. We introduce \textbf{MUSES}, a million-instance benchmark for prospective intellectual-roots retrieval over a fixed 2.33M-paper corpus, with roughly 140K test instances per familiarity tier. To our knowledge, it is the first prospective benchmark at this scale with a shared retrieval task and author-confirmed paper-level root labels. Alongside it, \textbf{CiteRoots} pairs a scalable rhetorical layer over local citation text (LLM judge $κ= 0.896$ versus human gold) with a paper-level author-endorsed layer ($n = 1{,}518$ generative-inspiration pairs from 753 focal papers). MUSES organizes difficulty along two axes: a \emph{familiarity} axis spanning CiteNext, CiteNew, and CiteNew-Isolated, and a \emph{functional} axis spanning broad citations, rhetorical roots, and author-endorsed roots. Across 9 method classes, a lean multi-centroid retriever built on SPECTER2 is strongest. Hit@100 falls from 0.534 on CiteNext to 0.424 on CiteNew, 0.205 on rhetorical CiteNew, and 0.171 on author-endorsed CiteNew, a $3.1\times$ decline. In a registered eight-lens full-test audit, roughly half of broad-tier test instances remain unsolved at K=1{,}000. Rhetorical role and author endorsement are distinct: the same judge agrees with endorsement at $κ= 0.037$. We release MUSES, both CiteRoots layers, and a distilled open companion judge for future work on prospective retrieval and intellectual roots.

cs.IR↗

OverdoseMoE: A Multi-Expert Framework for Opioid Overdose Risk Prediction

Opioid overdose remains a major clinical and public health burden, highlighting the need for scalable approaches to identify patients at high risk. Here, we investigate diagnosis-specific adaptation for 180-day opioid overdose risk prediction from patients' preceding one-year longitudinal ICD histories. We develop OODMAMBA and OODQWEN through continued pretraining on longitudinal diagnostic sequences followed by task-specific fine-tuning. Building on the stronger Qwen-based predictors, we further propose OVERDOSEMOE, a multi-expert framework that integrates models of different scales using complementary expert-weighting strategies. Diagnosis-specific adaptation consistently improved predictive performance over general-purpose language-model baselines, with OODQWEN achieving an AUPRC of 24.47 and an AUROC of 68.56. OVERDOSEMOE further improved discrimination and precision, achieving an AUPRC of 25.17 and an AUROC of 69.49 while outperforming the strongest single-model baselines. Among patients ranked in the top 5% of predicted risk, OVERDOSEMOE identified substantially enriched overdose risk, achieving a PPV of 25.38% while retaining meaningful recall. Evaluation on an independent MIMIC-IV cohort further demonstrated cross-cohort robustness, with complementary weighting strategies showing advantages across different performance measures. These findings demonstrate that diagnosis-specific language-model adaptation combined with multi-expert integration can improve opioid overdose risk stratification and support more robust prediction across heterogeneous electronic health record populations.

cs.CL↗

The Refutation Gap: Certifying Both Halves of an Optimality Claim

Synthesis pipelines increasingly claim not just that a program is correct, but that it is optimal. Such a claim has two halves with radically different verification stories. The upper bound, "a program of size m exists", is witnessed by an artifact that can be re-executed, proved equivalent to its specification, and shipped with a machine-checked certificate. The lower bound, "no program of size m-1 exists", has no witness and is discharged by running a solver until it reports UNSAT. Combinatorial optimization has known this asymmetry for decades and has largely addressed it: certifying algorithms make it explicit (McConnell et al., 2011), and pseudo-Boolean proof logging can certify optimality end to end with a formally verified checker (Bogaerts et al., 2023; Koops et al., 2025). That discipline has not reached circuit minimization. We call this the refutation gap: published gate counts for minimal XOR circuits provide no certificate for either half of the claim, and neither did 121 optimality results we ourselves produced. We close the gap with a pipeline that synthesizes minimal linear straight-line programs over GF(2), where every decisive UNSAT answer emits a DRAT proof checked by an independent third-party checker. We certify all 121 optimality results established by the project, across n = 6 to 9: 111 carry independently verified refutations, 10 are closed by a free counting bound, and none disagrees with the uncertified value. The median proof is 1.1 MB and checking costs 1.9x solving. We give five case studies where verification caught defects that code review did not, report two interface obstacles that push practitioners toward the uncertified path, and describe an adversarial audit that revealed a failure tail we were about to attribute to the problem was actually caused by our own budget.

cs.LO↗

Demons on a Budget: Adaptive Measurement Placement at the Entanglement Phase Transition

Monitored quantum circuits exhibit a measurement-induced phase transition between volume-law and area-law entanglement as a function of the measurement rate $p$. Prior work places measurements at random locations and treats the rate as the control parameter. We instead fix the measurement budget and vary the placement process, comparing random placement against hand-designed and learned policies in brickwork random Clifford circuits at matched budget. First, placement geometry matters more than placement information. A deterministic contiguous sweep cuts the half-cut entropy by a factor of 3.4 relative to random placement, while equal-coverage unstructured placement and a greedy policy with full state access do far worse. The effect is carried by spatial order alone: measuring the $k$ least recently measured sites gives $4.14 \pm 0.06$ bits with random tie-breaking and $1.29 \pm 0.04$ bits with position-ordered tie-breaking. Second, the sweep eliminates the transition rather than shifting it. Tripartite mutual information crossings recede as $p^* \propto 1/L$, the steady-state entropy saturates at an $L$-independent ceiling near $0.46/p$, and data for $64 \le L \le 512$ collapse onto the form $S = p^{-1} f(pL)$ predicted by a ballistic regrowth argument. Third, in stabilizer dynamics every outcome is deterministic or a fair coin flip, so the record's Shannon entropy is exactly countable; the sweep dominates the entropy-versus-record-cost frontier while paying the same roughly one bit per measurement as random placement. Policies trained by cross-entropy and proximal policy optimization do not find the sweep: score-based policies parameterize which sites to measure, not the order in which degenerate scores are resolved, and the effect lives in that order. The phase diagram of monitored dynamics is a property of the placement process, not only of the measurement rate.

quant-ph↗

Non-Definability of Reachability in Büchi Arithmetic for a Family of Generalized Collatz Maps

Let $q \ge 3$ and $d \ge 1$ be odd integers with $q+d$ a power of $2$. We study the generalized Collatz map $T_{q,d}$, a one-dimensional piecewise-affine map on the positive integers, and its unparameterized reachability relation $R(x,z)$, which holds when $z$ is an iterate of $x$ under $T_{q,d}$. We prove that for every such pair $(q,d)$ the relation $R$ is not first-order definable in Büchi arithmetic $\langle \mathbb{N}, +, V_q \rangle$. Equivalently, no finite automaton recognizes the base-$q$ encoding of $R$. Assuming definability of $R$, we construct a first-order formula that defines the set of powers of $2$. Cobham's theorem then rules out this set. The family includes the classical map $T_{3,1}$. The family is infinite but restricted, isolated by the condition that $q+d$ is a power of $2$ . Unlike the undecidability results of Conway, Kurtz, and Simon, the construction does not embed universal computation and does not depend on the Collatz conjecture.

math.GM↗

FactorLibrary: From Polynomials to Circuits via Recursive Subgoals

Finding minimal arithmetic circuits for polynomials over finite fields is a combinatorially hard problem central to algebraic complexity theory. We formulate it as a reinforcement learning problem in two directions, bottom-up and top-down. To address the challenge of a fast-growing combinatorial search space, we introduce FactorLibrary, which stores factorizable subexpressions that serve as reusable subgoals across training episodes. We trained a bottom-up agent with Gumbel-PPO-MCTS and two top-down agents with PPO+MCTS and SAC. The PPO+MCTS top-down agent exhibited the most stable performance, finding certified optimal circuits up to complexity $8$ with a success rate of $91.8\%$.

cs.LG↗

Logical Undefinability of the Generalized Collatz Transition Relation in Büchi Arithmetic

Let $q$ be an odd prime and let $d$ be an odd integer. We show that the arbitrary-step transition relation of the generalized Collatz map $T_{q,d}$ is not first-order definable in Base-2 Büchi Arithmetic ($BA_2$). We do this by demonstrating that if the transition relation were definable, the exponential set $P_q = \{q^y : y \in \mathbb{N}\}$ would also be definable in $BA_2$. Since $P_q$ is strictly non-semilinear, this yields a direct contradiction with the Cobham--Semënov theorem. Consequently, we demonstrate that no finite automaton reading base-2 representations can recognize this transition relation.

math.NT↗

Failure Modes of Deep Multi-Agent RL in Asynchronous Pricing: Reproducible Triggers, Trace Diagnostics, and a Partial Fix

We study two reproducible failure modes of deep multi-agent reinforcement learning in continuous-time pricing markets: (i) tacit cartel formation between competing DDPG agents, and (ii) actor--critic instability at high event rates. We instantiate both inside a single CT-MARL benchmark (Poisson-clocked price updates, observation latency $δ$, interior-optimum logit demand), show that synchronous DDPG agents reliably trigger Failure Mode 1 with collusion index $Δ= 0.69 \pm 0.11$, and quantify a partial microstructure fix: asynchrony alone cuts collusion by 48\% and adding latency drives it to a minimum of $Δ= 0.28$. The fix has clearly documented costs: it is partial ($Δ$ remains supra-Bertrand), it is non-monotone in $δ$, and it does not survive Failure Mode 2, which emerges as DDPG critic divergence at $λ= 5$ and corrupts the phase-diagram cell at $(λ{=}5, δ{=}1)$. We accompany the scalar collusion index with trajectory-level trace diagnostics that expose the within-episode signalling collapse and the post-shock non-recovery.

cs.MA↗

ThinkSwitch: Context Distillation with LoRA and Weight Interpolation for Specific-Purpose Reasoning Tasks

Large language models often improve on difficult tasks by spending inference-time compute on a reasoning trace before producing the final answer. That extra computation can be useful, but it also raises latency, token cost, and deployment complexity. We introduce \textbf{ThinkSwitch}, a low-compute procedure for co-training paired instruct and thinking checkpoints. Starting from compatible Qwen3-4B instruct and thinking models, each iteration asks the thinking checkpoint to generate answers, removes the reasoning trace, distills the answer-only pairs into the instruct checkpoint with QLoRA, and reconstructs a thinking checkpoint with spherical weight interpolation. The only human-supplied inputs are task prompts; the labels are generated by the model itself. On a 30-question AIME 2026 evaluation, ThinkSwitch improves the instruct checkpoint from 10/30 to 20/30 and the thinking checkpoint from 14/30 to 22/30. On a 30-question PubMedQA subset, it improves the instruct checkpoint from 13/30 to 18/30 and the thinking checkpoint from 18/30 to 25/30. The complete experiment uses 15 training prompts per domain and costs \$2.86 on a single cloud RTX 3070. The results are small-scale, but they indicate that targeted distillation loops can move part of the benefit of explicit reasoning into weights while preserving a separate thinking mode.

cs.LG↗

Poisoning the Watchtower: Prompt Injection Attacks Against LLM-Augmented Security Operations Through Adversarial Log Content

Large language models (LLMs) are increasingly used as analyst assistants in security operations centers (SOCs), where they ingest log and alert data to produce triage labels, incident summaries, or remediation advice. We study a structural failure mode of this design: many log fields are attacker controlled. User agents, URLs, payloads, DNS queries, and attempted usernames can therefore carry instructions to the model alongside evidence of the intrusion. We call this setting \emph{log-substrate prompt injection}. We introduce a four-class taxonomy of log-substrate attacks: direct override (S1), persona hijack (S2), context manipulation (S3), and obfuscated payloads (S4). We evaluate 48 strategy-defense-task combinations using \texttt{gpt-4o-mini} as the analyst. Three findings stand out. First, direct overrides are ineffective in our setting: all S1 classification attacks achieve 0\% suppression. In contrast, persona hijacks suppress 68\% of malicious logs under a naive classifier and remain effective under stronger defenses. Second, summarization is the highest-risk task: context manipulation reaches 96\% injection success without defenses and 38\% even with constrained output. Third, defenses reduce but do not eliminate the attack surface: average injection success falls from 26.6\% under naive prompting to 11.8\% under our strongest defense. We also compare empirical results to a deterministic mock analyst and find that simulation substantially mispredicts current model behavior, especially for direct overrides. These results suggest that SOC copilots should treat raw log content as adversarial input rather than ordinary analyst context.

cs.CR↗

Beyond the Answer: Decoding the Behavior of LLMs as Scientific Reasoners

As Large Language Models (LLMs) achieve increasingly sophisticated performance on complex reasoning tasks, current architectures serve as critical proxies for the internal heuristics of frontier models. Characterizing emergent reasoning is vital for long-term interpretability and safety. Furthermore, understanding how prompting modulates these processes is essential, as natural language will likely be the primary interface for interacting with AGI systems. In this work, we use a custom variant of Genetic Pareto (GEPA) to systematically optimize prompts for scientific reasoning tasks, and analyze how prompting can affect reasoning behavior. We investigate the structural patterns and logical heuristics inherent in GEPA-optimized prompts, and evaluate their transferability and brittleness. Our findings reveal that gains in scientific reasoning often correspond to model-specific heuristics that fail to generalize across systems, which we call "local" logic. By framing prompt optimization as a tool for model interpretability, we argue that mapping these preferred reasoning structures for LLMs is an important prerequisite for effectively collaborating with superhuman intelligence.

cs.AI↗

CircuitBuilder: From Polynomials to Circuits via Reinforcement Learning

Motivated by auto-proof generation and Valiant's VP vs. VNP conjecture, we study the problem of discovering efficient arithmetic circuits to compute polynomials, using addition and multiplication gates. We formulate this problem as a single-player game, where an RL agent attempts to build the circuit within a fixed number of operations. We implement an AlphaZero-style training loop and compare two approaches: Proximal Policy Optimization with Monte Carlo Tree Search (PPO+MCTS) and Soft Actor-Critic (SAC). SAC achieves the highest success rates on two-variable targets, while PPO+MCTS scales to three variables and demonstrates steady improvement on harder instances. These results suggest that polynomial circuit synthesis is a compact, verifiable setting for studying self-improving search policies.

cs.LG↗

Predicting first-episode homelessness among US Veterans using longitudinal EHR data: time-varying models and social risk factors

Homelessness among US veterans remains a critical public health challenge, yet risk prediction offers a pathway for proactive intervention. In this retrospective prognostic study, we analyzed electronic health record (EHR) data from 4,276,403 Veterans Affairs patients during a 2016 observation period to predict first-episode homelessness occurring 3-12 months later in 2017 (prevalence: 0.32-1.19%). We constructed static and time-varying EHR representations, utilizing clinician-informed logic to model the persistence of clinical conditions and social risks over time. We then compared the performance of classical machine learning, transformer-based masked language models, and fine-tuned large language models (LLMs). We demonstrate that incorporating social and behavioral factors into longitudinal models improved precision-recall area under the curve (PR-AUC) by 15-30%. In the top 1% risk tier, models yielded positive predictive values ranging from 3.93-4.72% at 3 months, 7.39-8.30% at 6 months, 9.84-11.41% at 9 months, and 11.65-13.80% at 12 months across model architectures. Large language models underperformed encoder-based models on discrimination but showed smaller performance disparities across racial groups. These results demonstrate that longitudinal, socially informed EHR modeling concentrates homelessness risk into actionable strata, enabling targeted and data-informed prevention strategies for at-risk veterans.

cs.CL↗

Parity-Dependent Real-Rootedness in Independence Polynomials of Generalized Petersen Graphs

We investigate the distribution of zeros of the independence polynomial ${\rm I}(G, x)$ for the family of Generalized Petersen graphs ${\rm GP}(n, k)$ in the complex plane. While the independence numbers and coefficients of these graphs have been studied, the global behavior of their roots remains largely unexplored. Using an exact transfer matrix algorithm parameterized by $k$, we compute ${\rm I}({\rm GP}(n,k), x)$ for $n$ up to $30$ and $k \in \{1, 2, 3, 4\}$. Our numerical analysis reveals a striking parity-based dichotomy: for odd $k$, the roots exhibit complex conjugate structures accumulating on closed curves, whereas for even $k$, the roots appear to be strictly real and negative. Motivated by this evidence, we conjecture that ${\rm I}({\rm GP}(n,k), x)$ is real-rooted, and hence log-concave, if and only if $k$ is even. This phenomenon connects algebraic properties of ${\rm GP}(n,k)$ to questions about zero-free regions and limiting behavior in the hard-core lattice gas model.

math.CO↗

Amortized Inference for Model Rocket Aerodynamics: Learning to Estimate Physical Parameters from Simulation

Accurate prediction of model rocket flight performance requires estimating aerodynamic parameters that are difficult to measure directly. Traditional approaches rely on computational fluid dynamics or empirical correlations, while data-driven methods require extensive real flight data that is expensive and time-consuming to collect. We present a simulation-based amortized inference approach that trains a neural network on synthetic flight data generated from a physics simulator, then applies the learned model to real flights without any fine-tuning. Our method learns to invert the forward physics model, directly predicting drag coefficient and thrust correction factor from a single apogee measurement combined with motor and configuration features. In this proof-of-concept study, we train on 10,000 synthetic flights and evaluate on 8 real flights, achieving a mean absolute error of 12.3 m in apogee prediction - demonstrating promising sim-to-real transfer with zero real training examples. Analysis reveals a systematic positive bias in predictions, providing quantitative insight into the gap between idealized physics and real-world flight conditions. We additionally compare against OpenRocket baseline predictions, showing that our learned approach reduces apogee prediction error. Our implementation is publicly available to support reproducibility and adoption in the amateur rocketry community.

cs.LG↗

Quantization Blindspots: How Model Compression Breaks Backdoor Defenses

Backdoor attacks embed input-dependent malicious behavior into neural networks while preserving high clean accuracy, making them a persistent threat for deployed ML systems. At the same time, real-world deployments almost never serve full-precision models: post-training quantization to INT8 or lower precision is now standard practice for reducing memory and latency. This work asks a simple question: how do existing backdoor defenses behave under standard quantization pipelines? We conduct a systematic empirical study of five representative defenses across three precision settings (FP32, INT8 dynamic, INT4 simulated) and two standard vision benchmarks using a canonical BadNet attack. We observe that INT8 quantization reduces the detection rate of all evaluated defenses to 0% while leaving attack success rates above 99%. For INT4, we find a pronounced dataset dependence: Neural Cleanse remains effective on GTSRB but fails on CIFAR-10, even though backdoors continue to survive quantization with attack success rates above 90%. Our results expose a mismatch between how defenses are commonly evaluated (on FP32 models) and how models are actually deployed (in quantized form), and they highlight quantization robustness as a necessary axis in future evaluations and designs of backdoor defenses.

cs.LG↗

Analyzing the birth-death model of Oncostreams in Glioma, and the effects of Cytochalasin D treatment

This research project investigates the critical role of oncostreams in glioma aggressiveness, leveraging advanced ex-vivo 3D explants and in-vivo intravital imaging techniques to establish a direct correlation between oncostream density and cancer severity. The primary objective is to model the cell populations within oncostreams, with a specific focus on GFP+ NPA cells, to simulate cancer dynamics and provide insights into tumor behavior. The study employs a simple Birth-Death process to analyze cell population dynamics and treatment effects, building and solving Kolmogorov equations to predict changes in cell population over time. While the model could be expanded to include additional modulators such as morphological attributes and neurotransmitter exposure, the focus remains on cell population to maintain feasibility. The study also examines various treatment methods, finding that glutamate increases glioma cell movement while histamine reduces it. Collagenase treatment effectively dismantles oncostreams, suggesting a potential therapeutic strategy. For this paper, we specifically are going to be looking at Cytochalasin D, which shows promise in disrupting oncostreams and reducing glioma invasiveness. By integrating these treatment variables into the model, the research aims to understand their impact on glioma cell density within the oncostreams and aggressiveness, thereby contributing to improved cancer management strategies. This comprehensive approach is expected to enhance our understanding of glioma progression and inform the development of effective therapeutic interventions.

q-bio.TO↗