arXiv subjects

Aditya Banerjee

Publications and source records attributed to Aditya Banerjee.

9 recordsLinked to original sources

arXiv2026

TraceSQL: Traceable Answerability Estimation for Reference-Free Text-to-SQL Verification

Text-to-SQL systems are commonly evaluated using ground-truth SQL queries or reference execution results, but such supervision is unavailable at inference time in real-world deployments. This creates a critical verification problem: given only a user question, database context, and generated SQL, can a system estimate whether the generated query is likely to correctly answer the question? Recent approaches use LLMs as judge or specialized agents to inspect generated SQL, but their decisions can be difficult to trace. Outcome Reward Models (ORMs) address this by learning from execution-labeled candidate SQLs and assigning correctness scores to unseen queries, yet they still provide limited visibility into the signals behind each verification. To address this limitation, we propose TraceSQL, a lightweight and traceable verification model built on explicit diagnostic features. TraceSQL combines 67 features capturing question ambiguity, question requirements, question-schema-SQL consistency, SQL structure, and intent alignment. These signals remain available for examining which factors influence each prediction and for tracing decisions back to diagnostic evidence. On BIRD development databases, TraceSQL achieves 66.47% F1 and 64.48% ROC-AUC, compared with 61.87% F1 and 58.26% ROC-AUC for the GradeSQL-7B ORM baseline on the same generated-SQL evaluation. Feature attribution further shows that the model relies on both semantic grounding and deterministic SQL-structure signals. These results show that SQL verification can be performed with a lightweight learned model while retaining feature-level evidence for inspecting and diagnosing its predictions.

cs.CL↗

arXiv2026

Trainable Nonexpansive Denoisers for Contractive Image Reconstruction

Trainable denoisers with Lipschitz control have become central to convergent image reconstruction. However, training neural networks that simultaneously offer strong denoising performance and global Lipschitz guarantees is challenging. Existing approaches enforce Lipschitz control only empirically, providing no guarantees beyond the training data. In this work, we show that by exploiting the action of permutations on the image lattice, we can constrain a neural architecture that is globally nonexpansive (Lipschitz bound $\leqslant 1$). We integrate the proposed denoiser with forward imaging operators to develop a reconstruction mechanism that is provably contractive and therefore globally convergent. Experiments on standard inverse problems, such as superresolution and deblurring, demonstrate that our reconstruction performance is competitive with softly constrained baselines while providing Lipschitz guarantees.

eess.IV↗

arXiv2026

SemJoin: Semantic Join Optimization

Integrating unstructured data into relational database systems is increasingly important as demand grows for natural language querying and analysis. A semantic join, joining two tables under a natural-language predicate, can be evaluated with a large language model (LLM), but comparing every pair of tuples requires O(M x N) LLM invocations and is cost-prohibitive at scale. Existing systems reduce this cost but typically commit to a single fixed strategy (e.g., embedding similarity or one batched scheme) regardless of the data or the join predicate. We propose an LLM-agent-based decision pipeline that optimizes semantic joins by matching the execution strategy to the characteristics of the underlying tables. An LLM advisor routes each join to one of two strategies: a Cluster Join, which prunes candidates via unsupervised embedding clustering and sample-based filtering, or a Classifier strategy for predicates that reduce to a shared discrete label set. Across three diverse datasets (IMDb reviews, email contradictions, and Stack Overflow tags), the advisor consistently identifies the optimal execution strategy for each workload. This dynamic routing proves decisive: it outperforms adaptive block join (ABJ) by 20-33 F1 points across all datasets while consuming fewer tokens on two of the three, and achieves higher F1 scores than featurized-decomposition join (FDJ) at one to two orders of magnitude lower token cost.

cs.DB↗

arXiv2025

Quantum scarring enhances non-Markovianity of subsystem dynamics

Given that any subsystem of a closed out-of-equilibrium quantum system is an open quantum system, its dynamics (reduced from the full system's unitary evolution) can be either Markovian (memory-less) or non-Markovian, with the latter necessarily impeding the process of relaxation and thermalization. Seemingly independently, such non-ergodic dynamics occurs when an initial state has spectral weight on the so-called quantum many-body scar states, which are non-thermalizing eigenstates embedded deep in the spectrum of otherwise thermal eigenstates. In this article, we present numerical evidence that, in the class of systems which exhibit scars-induced entanglement oscillations, the presence of quantum scars is a microscopic ingredient that enables and enhances non-Markovianity of the dynamics of subsystems. We exemplify this with the PXP model and its deformations which either enhance or erase the signatures of scarred dynamics when quenched from simple product states with significant overlaps with the scarred states. The effect of thermalizing or scarring initial states is also similarly investigated. By probing information backflows with the dynamical behaviour of the distances between temporally-separated transient states of small subsystems, systematic signatures of subsystem non-Markovianity in these models are presented. It is seen that scarring-enhancing (erasing) deformations also exhibit enhanced (diminished) subsystem non-Markovianity. Likewise, results relating scarring (thermalizing) initial states to stronger (weaker) subsystem non-Markovianity are also presented. The retention of memory and revivals between transient subsystem states is a finer form of memory effect than captured by the revivals of full system's fidelity with the initial states. This sheds new light on the dynamical memories associated with quantum scarring (abstract shortened due to arxiv limitations).

quant-ph↗

arXiv2025

Fine-grained dynamics of entanglement in non-integrable quenches far across the Ising quantum critical point

The task of exploring and understanding various aspects of far-from-equilibrium dynamics of closed and generic quantum many-body systems has received a thrust of attention in recent years, driven partly by remarkable advances in ultracold experimental technologies. In this work, for the paradigmatic Ising spin chain with transverse and longitudinal fields and partly motivated by the practice of site-resolved control in contemporary ultracold experiments, we present numerical observations of several $\textit{fine-grained}$ (small-subsystem level) features of far-from-equilibrium dynamics from a quantum informational point of view, induced by quantum quenches far across the Ising critical point between states deep inside the para- and ferro-magnetic regimes. Rather featureless dynamics is seen for ferromagnetic to paramagnetic quenches, but paramagnetic to ferromagnetic quenches exhibit rich behaviour, including recurrences of an approximately Page-like dynamics of entanglement entropies of one- and two-spin subsystems, periodic but short-lived occurrences of approximately $1-$uniform states, a series of sudden deaths and revivals of entanglement between two spins in the system's bulk, non-analytic cusps in single-copy entanglement entropy for three-spin and bigger subsystems, insufficient mixedness and a series of scrambling-$\textit{un}$scrambling of local mutual information between neighboring spins. Moreover, essentially indistinguishable dynamics is seen at very early times between the integrable limit (zero longitudinal field) and non-integrable cases, with the former eventually showing signatures of better mixing and faster approach to equilibration than the latter. These features are expected to hold for quench dynamics across Ising quantum critical points in more complicated systems.

quant-ph↗

arXiv2025

Non-Markovianity of subsystem dynamics in isolated quantum many-body systems

It is believed that an isolated and far-from-equilibrium quantum many-body system should try to attain equilibrium via a mechanism whereby any given subsystem acts as an open quantum system that is coupled to an environment, which is the complementary part of the full system, and undergoes a complicated equilibration process such that all the subsystems in the long-time limit attain equilibrium states compatible with the global equilibrium state. This picture begs the question whether the dynamics of any given subsystem is Markovian (monotonic loss of information and memory) or non-Markovian. In this work, by numerically probing the dynamical behaviour of the quantum distances between $\textit{temporally-separated}$ states of small subsystems, we reveal the telltale signatures of (non-)Markovianity of the dynamics of subsystems of an isolated quantum spin system brought in the far-from-equilibrium regime, exemplified with the mixed-field Ising spin chain quenched between parameter regimes deep inside its magnetically ordered and disordered regimes. Additionally, remarkably systematic behaviour is seen in a measure of classical distances between the quantum states of the considered subsystems. These features strongly depend on the direction of quenching in the parameter space, with paramagnetic-to-ferromagnetic quenches offering considerably stronger signatures of subsystem non-Markovianity, for which we offer heuristic arguments.

quant-ph↗

arXiv2021

Entanglement Spectrum in General Free Fermionic Systems

The statistical mechanics characterization of a finite subsystem embedded in an infinite system is a fundamental question of quantum physics. Nevertheless, a full closed form { for all required entropic measures} does not exist in the general case even for free systems when the finite system in question is composed of several disjoint intervals. Here we develop a mathematical framework based on the Riemann-Hilbert approach to treat this problem in the one-dimensional case where the finite system is composed of two disjoint intervals and in the thermodynamic limit (both intervals and the space between them contains an infinite number of lattice sites and the result is given as a thermodynamic expansion). To demonstrate the usefulness of our method, we compute the change in the entanglement and negativity namely the spectrum of eigenvalues of the reduced density matrix with our without time reversal of one of the intervals. We do this in the case that the distance between the intervals is much larger than their size. The method we use can be easily applied to compute any power in an expansion in the ratio of the distance between the intervals to their size. {We expect these results to provide the necessary mathematical apparatus to address relevant questions in concrete physical scenarios, namely the structure and extent of quantum correlations in fermionic systems subject to local environment.

quant-ph↗

arXiv2021

Modular Procedural Generation for Voxel Maps

Task environments developed in Minecraft are becoming increasingly popular for artificial intelligence (AI) research. However, most of these are currently constructed manually, thus failing to take advantage of procedural content generation (PCG), a capability unique to virtual task environments. In this paper, we present mcg, an open-source library to facilitate implementing PCG algorithms for voxel-based environments such as Minecraft. The library is designed with human-machine teaming research in mind, and thus takes a 'top-down' approach to generation, simultaneously generating low and high level machine-readable representations that are suitable for empirical research. These can be consumed by downstream AI applications that consider human spatial cognition. The benefits of this approach include rapid, scalable, and efficient development of virtual environments, the ability to control the statistics of the environment at a semantic level, and the ability to generate novel environments in response to player actions in real time.

cs.AI↗

arXiv2020

Relevance in the Renormalization Group and in Information Theory

The analysis of complex physical systems hinges on the ability to extract the relevant degrees of freedom from among the many others. Though much hope is placed in machine learning, it also brings challenges, chief of which is interpretability. It is often unclear what relation, if any, the architecture- and training-dependent learned "relevant" features bear to standard objects of physical theory. Here we report on theoretical results which may help to systematically address this issue: we establish equivalence between the information-theoretic notion of relevance defined in the Information Bottleneck (IB) formalism of compression theory, and the field-theoretic relevance of the Renormalization Group. We show analytically that for statistical physical systems described by a field theory the "relevant" degrees of freedom found using IB compression indeed correspond to operators with the lowest scaling dimensions. We confirm our field theoretic predictions numerically. We study dependence of the IB solutions on the physical symmetries of the data. Our findings provide a dictionary connecting two distinct theoretical toolboxes, and an example of constructively incorporating physical interpretability in applications of deep learning in physics.

cond-mat.stat-mech↗