arXiv subjects

Kaustav Mukherjee

Publications and source records attributed to Kaustav Mukherjee.

12 recordsLinked to original sources

arXiv2026

Spectral-Target Physical Latent Structuring for JEPA-Style World Models

Latent world models have become increasingly popular as a method to predict and plan in latent space rather than pixel space. Recent architectures, such as LeWorldModel (LeWM), jointly train the encoder and predictor using regularization techniques like SIGReg to prevent representation collapse. Even with such regularization preventing representation collapse, we identify a new world model failure mode of physical representation laziness, particularly noted in highly dynamic environments. For these lazy cases, the learned latent states do not collapse but nonetheless fail to represent key physical properties, causing ubiquitous downstream planning failure. To resolve this issue, we propose training-time auxiliary supervision with a lightweight "Fourier auxiliary head", which enforces physically-informed structuring of the latent space with no additional inference-time cost and can be generalized to any environment. Experimentally, we show that the auxiliary head substantially improves planning success rates in dynamic environments where the baseline LeWM exhibits physical representation laziness. It also leads to modest improvements in other environments, even when the baseline does not exhibit physical representation laziness. We further observe superior planning performance being accompanied by higher latent space correlations with key physical properties, indicating both the ability of our method to physically structure latent states and the potential planning-side benefit to the learned representation being physically structured. We also see in low-data regimes, auxiliary supervision is particularly impactful in increasing success rate. These findings support the use of our Fourier auxiliary head method to improve both overall success rate and data efficiency, while avoiding representation laziness in latent world models.

cs.LG↗

arXiv2026

Multi-Task Learning for Non-Canonical Phoneme Recognition via Articulatory Feature Decomposition

Pathological and more broadly non-canonical speech present significant challenges for automatic phoneme recognition due to systematic deviations from canonical pronunciation and limited availability of labeled clinical speech data. Existing phoneme recognition systems are typically trained on canonical speech and treat phonemes as atomic categorical labels, limiting their ability to detect structured articulatory errors common in speech disorders and accents. In this work, we introduce a linguistically structured approach to non-canonical phoneme recognition that decomposes phoneme prediction into articulatory feature dimensions such as manner, place, and voicing. We implement this formulation using a hierarchical multi-task learning architecture in which task-specific articulatory feature heads learn feature-level representations that are subsequently integrated through a cross-attention-based fusion module to produce phoneme predictions. To address the scarcity and noise of pathological speech labels, we combine this framework with semi-supervised learning via Momentum Pseudo-Labeling (MPL) and propose a cascaded training strategy that progressively introduces articulatory feature tasks while employing staged unfreezing of a pretrained speech encoder. Experiments on L2-ARCTIC, used as a proxy for pathological speech variation, show that the proposed approach achieves substantial improvements in phoneme recognition performance compared to strong baseline architectures, while yielding interpretable error patterns aligned with phonological feature structure. These results suggest that articulatory feature supervision is a promising strategy for robust and interpretable phoneme recognition in non-canonical speech, and motivate future validation on clinically diagnosed pathological speech datasets.

cs.SD↗

arXiv2026

Quantum simulation of Motzkin spin chain with Rydberg atoms

Motzkin spin chain is a well-known mathematical model with connections to symmetry-protected topological phases, such as the Haldane phase, as well as to concepts in the AdS/CFT correspondence. They exhibit highly entangled ground states that violate the area law and are exceptionally difficult to simulate with conventional numerical methods. Numerical simulations of the Motzkin ground state become further challenging at large system sizes due to their high-dimensional spin structure, rendering it a natural test bed for quantum simulation with ultra-cold systems. Here, we propose a Rydberg-atom based quantum simulation scheme that effectively realizes Motzkin spins using an experimentally accessible set of parameters. We show that the resulting effective Motzkin ground state reproduces the characteristic entanglement scaling and the block-structure properties of the reduced density matrix associated with the ideal Motzkin state. Our results establish a pathway toward a concrete experimental realization of Motzkin spins beyond purely mathematical constructions, opening avenues for exploring other similar exotic non-area-law entangled phases in programmable Rydberg simulators.

quant-ph↗

arXiv2025

Statistical learning on randomized data to verify quantum state approximate k-designs

Random ensembles of pure states have proven to be extremely important in various aspects of quantum physics such as benchmarking the performance of quantum circuits, testing for quantum advantage, providing novel insights for many-body thermalization and studying the black hole information paradox. Although generating a fully random ensemble is experimentally challenging, approximations of it are just as useful and are known to emerge naturally in a variety of physical models, including Rydberg setups. These are referred to as approximate quantum state designs, and verifying their degree of randomness can be an expensive task, similar to performing full quantum state tomography on many-body systems. In this theoretical work, we efficiently validate the character of approximate quantum designs with respect to data size acquisition when compared to the conventional frequentist approach. This is achieved by translating the information residing in the complex many-body state into a succinct representation of classical data using a random projective measurement basis, which is then processed using methods of statistical inference such as maximum likelihood estimation and neural networks and benchmarked against the predictions of shadow tomography. Our scheme of combining machine learning methods for postprocessing the data obtained from randomized measurements for efficient characterisation of (approximate) quantum state k designs is applicable to any noisy quantum platform that can generate quantum designs.

quant-ph↗

arXiv2025

Programmable glassy dynamics using tunable disorder in tweezer arrays

We propose a unifying framework for non-equilibrium relaxation dynamics in ensembles of positionally disordered interacting quantum spins based on the statistical properties, such as mean and variance, of the underlying disorder distribution. Our framework is validated through extensive exact numerical calculations and we use it to disentangle and understand the importance of dimensionality and interaction range for the observation of glassy (i.e., sub-exponential) decay dynamics. Leveraging the deterministic control of qubit positioning enabled by modern tweezer array architectures, we also introduce a method (``J-mapping'') that can be used to emulate the relaxation dynamics of a disordered system with arbitrary dimensionality and interaction range in bespoke one-dimensional arrays. Our approach paves the way towards tunable relaxation dynamics that can be explored in quantum simulators based on arrays of neutral atoms and molecules.

cond-mat.quant-gas↗

arXiv2024

Quantum network tomography of Rydberg arrays by machine learning

Configurable arrays of optically trapped Rydberg atoms are a versatile platform for quantum computation and quantum simulation, also allowing controllable decoherence. We demonstrate theoretically, that they also enable proof-of-principle demonstrations for a technique to build models for open quantum dynamics by machine learning with artificial neural networks, recently proposed in [Mukherjee et al. [arXiv:2409.18822] (2024)]. Using the outcome of quantum transport through a network of sites that correspond to excited Rydberg atoms, the multi-stage neural network algorithm successfully identifies the number of atoms (or nodes in the network), and subsequently their location. It further extracts an effective interaction Hamiltonian and decoherence operators induced by the environment. To probe the Rydberg array, one initiates dynamics repeatedly from the same initial state and then measures the transport probability to an output atom. Large datasets are generated by varying the position of the latter. Measurements are required in only one single basis, making the approach complementary to e.g. quantum process tomography. The cold atom platform discussed in this article can be used to explore the performance of the proposed protocol when training the neural network with simulation data, but then applying it to construct models based on experimental data.

quant-ph↗

arXiv2024

Automated quantum system modeling with machine learning

Despite the complexity of quantum systems in the real world, models with just a few effective many-body states often suffice to describe their quantum dynamics, provided decoherence is accounted for. We show that a machine learning algorithm is able to construct such models, given a straightforward set of quantum dynamics measurements. The effective Hilbert space can be a black box, with variations of the coupling to just one accessible output state being sufficient to generate the required training data. We demonstrate through simulations of a Markovian open quantum system that a neural network can automatically detect the number $N $ of effective states and the most relevant Hamiltonian terms and state-dephasing processes and rates. For systems with $N\leq5$ we find typical mean relative errors of predictions in the $10 \%$ range. With more advanced networks and larger training sets, it is conceivable that a future single software can provide the automated first stop solution to model building for an unknown device or system, complementing and validating the conventional approach based on physical insight into the system.

quant-ph↗

arXiv2024

Influence of disordered and anisotropic interactions on relaxation dynamics and propagation of correlations in tweezer arrays of Rydberg dipoles

We theoretically investigate the out-of-equilibrium dynamics of irregular one- and two-dimensional arrays of Rydberg dipoles featuring spatially anisotropic interactions. Starting from a collectively polarized initial state, we map out the dynamical phase diagram and identify a crossover between regimes of regular and anomalously slow relaxation of the initial collective order, that strongly depends on both the degree of interaction disorder and anisotropy. In addition, we find the regime of slow relaxation is characterized by a sub-ballistic propagation of correlations that remained confined to short distances even at long times. To explain our findings we develop an analytic model based on decoupled clusters of interacting dipoles that goes beyond prior theoretical works and enables us to identify multiple relaxation timescales. Our findings can be relevant for a wide variety of quantum science platforms naturally featuring disordered dipolar interactions, including polar molecules, frozen Rydberg gases and NV centers.

cond-mat.quant-gas↗

arXiv2022

Symmetry of surfaces for linear fractional group

We will compute the stable upper genus for the family of finite non-abelian simple groups $PSL_2(\mathbb{F}_p)$ for $p \equiv 3~(mod~4)$. This classification is well-grounded in the other branches of Mathematics like topology, smooth, and conformal geometry, algebraic categories.

math.CO↗

arXiv2022

Trapping and binding by dephasing

The binding and trapping of particles usually rely on conservative forces, described by unitary quantum dynamics. We show that both can also arise solely from spatially dependent dephasing, the simplest type of decoherence. This can be based on continuous weak position measurements in only selected regions of space, for which we propose a practical realisation. For a single particle, we demonstrate a quantum particle-in-the-box based on dephasing. For two particles, we demonstrate their binding despite repulsive interactions, if their molecular states are dephased at large separations only. Both mechanisms are experimentally accessible, as we show for an example with Rydberg atoms in a cold gas background.

quant-ph↗

arXiv2020

Two-dimensional spectroscopy of Rydberg gases

Two-dimensional (2D) spectroscopy uses multiple electromagnetic pulses to infer the properties of a complex system. A paradigmatic class of target systems are molecular aggregates, for which one can obtain information on the eigenstates, various types of static and dynamic disorder and on relaxation processes. However, two-dimensional spectra can be difficult to interpret without precise knowledge of how the signal components relate to microscopic Hamiltonian parameters and system-bath interactions. Here we show that two-dimensional spectroscopy can be mapped in the microwave domain to highly controllable Rydberg quantum simulators. By porting 2D spectroscopy to Rydberg atoms, we firstly open the possibility of its experimental quantum simulation, in a case where parameters and interactions are very well known. Secondly, the technique may provide additional handles for experimental access to coherences between system states and the ability to discriminate different types of decoherence mechanisms in Rydberg gases. We investigate the requirements for a specific implementation utilizing multiple phase coherent microwave pulses and a phase cycling technique to isolate signal components.

cond-mat.quant-gas↗

arXiv2005

Field induced first order antiferromagnetic to ferromagnetic transition in Al-doped CeFe$_2$: a calorimetric investigation

Field variation of heat capacity at fixed temperatures is investigated to identify the origin of the field induced first order phase transition in polycrystalline Ce(Fe$_{0.094}$Al$_{0.04}$)$_2$ sample. The heat capacity at 4.5K and 3.5K shows hysteresis in different field cycles and the virgin curve stays outside the envelope curve. This is analogous to the magnetization and magneto-resistance behavior observed in this system, where the amount of hysteresis, and the magnitude of zero-field irreversibility are attributed to the degree of supercooling/superheating, and the extent of kinetic arrest of the reverse transition from ferro- to antifrromagnetic state in field reducing cycle respectively. However, contrary to the magnetization and magneto-resistance, in heat capacity both these features have decreased with reducing temperature signifying the importance of structural contribution.

cond-mat.str-el↗