arXiv subjects

Soumya Ghosh

Publications and source records attributed to Soumya Ghosh.

At least 19 recordsLinked to original sources

arXiv2026

Strong Evidence for Formation of Hydroxyl Anion via 2-Particle-1-Hole Feshbach Resonances

This study investigates the formation of the hydroxyl anion (OH$^-$) via dissociative electron attachment in 2-propanol as a model system for studying both organic and inorganic molecules. Using high-level CAP-EOM-EA-CCSD calculations and advanced ToF mass spectrometry, we demonstrate that OH$^-$ formation at electron energies between 7 and 11 eV is dominated by two-particle-one-hole (2p-1h) Feshbach resonances. The potential-energy curves reveal a dense manifold of anionic states coupled through numerous avoided crossings, facilitating nonadiabatic population transfer during C-OH bond dissociation. Survival-probability analysis identifies a subset of six long-lived resonances that persist long enough to drive fragmentation, with states 25 and 28 acting as primary drivers by funneling the attached electron into the localized $\sigma^*(\mathrm{C{-}OH})$ antibonding orbital. These theoretical predictions are confirmed by experimental observations of a prominent OH$^-$ yield peaking at 8.6 eV, supporting a site-specific fragmentation mechanism that generalizes to the broader class of molecules.

physics.chem-ph↗

arXiv2026

Fermions on a 1D lattice: localized sources and sinks with dephasing

We study a general one-dimensional spinless fermionic system subject to a localized source, sink, and bulk dephasing. Within the Lindblad framework, we compute the time evolution of the density profile and spatial correlation functions. We find that the presence of bulk dephasing suppresses certain coherent quantum features, such as the Friedel oscillations, and it alters the transport dynamics to exhibit two distinct dynamical regimes instead of three as observed in the absence of dephasing. This effect can be understood as the destruction of ballistic motion caused by the dephasing noise. Under strong dephasing, the density profile becomes similar to the one expected for a classical diffusive regime. We also investigate the system in the presence of both a localized source and sink, placed at a distance of $\Delta$. Interestingly, quantum coherence generates secondary density peaks at integer multiples of the source-sink separation $\Delta$, which are systematically washed out as bulk dephasing drives the system toward the classical diffusive limit.

cond-mat.stat-mech↗

arXiv2026

Better Models, Faster Training: Sigmoid Attention for single-cell Foundation Models

Training stable biological foundation models requires rethinking attention mechanisms: we find that using sigmoid attention as a drop in replacement for softmax attention a) produces better learned representations: on six diverse single-cell datasets, sigmoid achieves 25% higher cell-type separation, better cell-type cohesion metrics, and lower validation loss, b) faster training, models with sigmoid attention train up to 10% faster than their softmax counterparts, and c) more stable training by eliminating inherent sources of instability in softmax attention. We establish that sigmoid attention has globally bounded derivatives ($\leq 0.25$) as opposed to softmax, and a diagonal Jacobian structure in contrast with softmax's dense coupling, which together help alleviate training instabilities. In stress tests on 160M-parameter bidirectional attention models trained without gradient clipping on 8K-token sequences, softmax diverges catastrophically, with gradients exploding by four orders of magnitude, while sigmoid remains stable. Finally, we implement and open-source TritonSigmoid, an efficient GPU kernel that achieves 515 TFLOPS on H100 GPUs, outperforming both FlashAttention-2 and FlashSigmoid, with native padding support, which is essential for biological sequences. Our results establish sigmoid attention as both theoretically grounded and empirically superior for biological foundation models. Code is available at https://github.com/MSDLLCpapers/triton-sigmoid

cs.LG↗

arXiv2026

Electron-Impact Quasi-Resonant Ion-Pair Dissociation of OCS: A Velocity Slice Imaging Study with Partial Wave Analysis

We present velocity map imaging data on intramolecular ion-pair dissociation (IPD) of carbonyl sulfide (OCS) induced by electron impact over the 20 eV to 45 eV energy range. Two distinct IPD pathways were resolved: CO+ + S- (threshold 14.8 +- 0.7 eV) and CS+ + O- (threshold 16.8 +- 0.7 eV). The kinetic energy release spectra display a single peak for S- but split into two components for O-; in both channels the maximum kinetic energies level off once the beam energy exceeds roughly 30 eV, pointing to excitation through discrete superexcited states of quasi-resonant character. Partial wave decomposition of the fragment angular distributions reveals that the momentum-transfer parameter (beta) surpasses unity at every energy studied, invalidating the dipole-Born approximation, and that the dominant partial wave character shifts systematically with beam energy. These patterns are consistent with a mechanism in which the incident electron deposits energy through inelastic scattering, populating hybrid Rydberg-ion-pair superexcited configurations that subsequently undergo state-specific unimolecular dissociation along nonadiabatic pathways. From an applied standpoint, intramolecular ion-pair dissociation matters for astrochemistry and radiation biophysics because it generates reactive anions and cations without photon emission, redistributing excess molecular energy nonadiabatically in environments ranging from interstellar clouds to biological systems.

physics.atm-clus↗

arXiv2026

Formation of Hydroxyl Anion via a 2-Particle 1-Hole Feshbach Resonance in DEA to 2-Propanol: A Joint Experimental and Theoretical Study

Absolute cross sections for the formation of OH- from 2-propanol (CH3CH(OH)CH3) via dissociative electron attachment (DEA) are reported in the incident electron energy range of 3.5-13 eV. Four fragment anions are observed: OH-, C2H2O-, C2H4O-, and C3H7O-. The OH- yield exhibits a pronounced resonance centered at 8.2 eV together with a broader structure extending over the 8-10 eV region. Equation-of-Motion Coupled-Cluster (electron attached) calculations with Singles and Doubles combined with a Complex Absorbing Potential (CAP/EOM-EA-CCSD) assign this feature to a two-particle-one-hole (2p-1h) core-excited Feshbach resonance. Potential energy curves along the C-OH dissociation coordinate reveal that core-excited anion states in this energy range promote efficient cleavage of the hydroxyl group. Analysis of Dyson orbitals and resonance widths demonstrates that only states with repulsive antibonding sigma(C-OH) character and sufficiently long lifetimes contribute significantly to the observed OH- production. These results provide fundamental insight into the DEA dynamics of secondary alcohols and highlight the role of multi-electron-attached resonances in site-specific bond rupture induced by low-energy electrons.

physics.atom-ph↗

arXiv2026

STRAND: Sequence-Conditioned Transport for Single-Cell Perturbations

Predicting how genetic perturbations change cellular state is a core problem for building controllable models of gene regulation. Perturbations targeting the same gene can produce different transcriptional responses depending on their genomic locus, including different transcription start sites and regulatory elements. Gene-level perturbation models collapse these distinct interventions into the same representation. We introduce STRAND, a generative model that predicts single-cell transcriptional responses by conditioning on regulatory DNA sequence. STRAND represents a perturbation by encoding the sequence at its genomic locus and uses this representation to parameterize a conditional transport process from control to perturbed cell states. Representing perturbations by sequence, rather than by a fixed set of gene identifiers, supports zero-shot inference at loci not seen during training and expands inference-time genomic coverage from ~1.5% for gene-level single-cell foundation models to ~95% of the genome. We evaluate STRAND on CRISPR perturbation datasets in K562, Jurkat, and RPE1 cells. STRAND improves discrimination scores by up to 33% in low-sample regimes, achieves the best average rank on unseen gene perturbation benchmarks, and improves transfer to novel cell lines by up to 0.14 in Pearson correlation. Ablations isolate the gains to sequence conditioning and transport, and case studies show that STRAND resolves functionally alternative transcription start sites missed by gene-level models.

q-bio.GN↗

arXiv2025

Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methods

Training data attribution (TDA) is concerned with understanding model behavior in terms of the training data. This paper draws attention to the common setting where one has access only to the final trained model, and not the training algorithm or intermediate information from training. We reframe the problem in this "final-model-only" setting as one of measuring sensitivity of the model to training instances. To operationalize this reframing, we propose further training, with appropriate adjustment and averaging, as a gold standard method to measure sensitivity. We then unify existing gradient-based methods for TDA by showing that they all approximate the further training gold standard in different ways. We investigate empirically the quality of these gradient-based approximations to further training, for tabular, image, and text datasets and models. We find that the approximation quality of first-order methods is sometimes high but decays with the amount of further training. In contrast, the approximations given by influence function methods are more stable but surprisingly lower in quality.

cs.LG↗

arXiv2025

Study of low-energy electron-induced dissociation of 1-Propanol

The fragmentation of 1-propanol resulting from dissociative electron attachment has been explored across an energy range of 3.5 to 16 eV. Four distinct ion species are identified: $\text{H}^{-}$, $\text{O}^{-}$, $\text{OH}^{-}$, and $\text{C}_{3}\text{H}_{7}\text{O}^{-}$. The $\text{OH}^{-}$ ion exhibited a prominent peak near 8.7 eV, along with a small hump near 5.6 eV. Complementary channels led to the formation of the $\text{H}^{-}$ and $\text{C}_{3}\text{H}_{7}\text{O}^{-}$ ions. Both these two ions exhibit a sharp peak near 6 eV and broad overlapping resonances between 7 to 12 eV. The observed ion yields of distinct dissociation fragments in this study, when compared with those from previously studied alcohols, suggest site-specific fragmentation of alcohols during dissociative electron attachment. To gain a deeper understanding of the dissociation pathways, Density Functional Theory~(DFT) calculations were conducted, revealing the threshold energies for each channel. These threshold energies aligned well with the experimental uncertainties.

physics.atm-clus↗

arXiv2025

When in Doubt, Cascade: Towards Building Efficient and Capable Guardrails

Large language models (LLMs) have convincing performance in a variety of downstream tasks. However, these systems are prone to generating undesirable outputs such as harmful and biased text. In order to remedy such generations, the development of guardrail (or detector) models has gained traction. Motivated by findings from developing a detector for social bias, we adopt the notion of a use-mention distinction - which we identified as the primary source of under-performance in the preliminary versions of our social bias detector. Armed with this information, we describe a fully extensible and reproducible synthetic data generation pipeline which leverages taxonomy-driven instructions to create targeted and labeled data. Using this pipeline, we generate over 300K unique contrastive samples and provide extensive experiments to systematically evaluate performance on a suite of open source datasets. We show that our method achieves competitive performance with a fraction of the cost in compute and offers insight into iteratively developing efficient and capable guardrail models. Warning: This paper contains examples of text which are toxic, biased, and potentially harmful.

cs.CL↗

arXiv2025

Multi-Level Explanations for Generative Language Models

Despite the increasing use of large language models (LLMs) for context-grounded tasks like summarization and question-answering, understanding what makes an LLM produce a certain response is challenging. We propose Multi-Level Explanations for Generative Language Models (MExGen), a technique to provide explanations for context-grounded text generation. MExGen assigns scores to parts of the context to quantify their influence on the model's output. It extends attribution methods like LIME and SHAP to LLMs used in context-grounded tasks where (1) inference cost is high, (2) input text is long, and (3) the output is text. We conduct a systematic evaluation, both automated and human, of perturbation-based attribution methods for summarization and question answering. The results show that our framework can provide more faithful explanations of generated output than available alternatives, including LLM self-explanations. We open-source code for MExGen as part of the ICX360 toolkit: https://github$.$com/IBM/ICX360.

cs.CL↗

arXiv2025

Large Language Model Confidence Estimation via Black-Box Access

Estimating uncertainty or confidence in the responses of a model can be significant in evaluating trust not only in the responses, but also in the model as a whole. In this paper, we explore the problem of estimating confidence for responses of large language models (LLMs) with simply black-box or query access to them. We propose a simple and extensible framework where, we engineer novel features and train a (interpretable) model (viz. logistic regression) on these features to estimate the confidence. We empirically demonstrate that our simple framework is effective in estimating confidence of Flan-ul2, Llama-13b, Mistral-7b and GPT-4 on four benchmark Q\&A tasks as well as of Pegasus-large and BART-large on two benchmark summarization tasks with it surpassing baselines by even over $10\%$ (on AUROC) in some cases. Additionally, our interpretable approach provides insight into features that are predictive of confidence, leading to the interesting and useful discovery that our confidence models built for one LLM generalize zero-shot across others on a given dataset.

cs.CL↗

arXiv2025

Electron Impact Fragmentation Dynamics of Carbonyl Sulfide: A Combined Experimental and Theoretical Study

In this study, we examine the interactions of low- to intermediate-energy electrons (0$-$45 eV) with carbonyl sulfide (OCS). These collisions lead to the formation of several anionic fragments, including C$^-$, O$^-$, S$^-$, and SO$^-$. When the incident electron energy is below the first ionization potential of the molecule, dissociative electron attachment (DEA) process dominates, primarily yielding O$^-$ and S$^-$ fragments. At higher energies, beyond the ionization potential, ion-pair dissociation (IPD) becomes the dominant process, resulting in the emergence of additional fragments such as C$^-$ and SO$^-$. This leads to an increasingly intricate mechanism, necessitating a detailed analysis to elucidate the ion-pair dissociation pathways. The absolute cross section for S$^-$ ions has been determined using the well-established relative flow technique. Theoretical cross sections are calculated using the multi-configurational time-dependent hartree (MCTDH) method, with each potential energy curve obtained from equation-of-motion coupled-cluster singles and doubles (EOM-CCSD) calculations. The computed values are in excellent agreement with the experimental data. The analysis reveals contributions from both linear and bent anionic resonant states. Due to low count rates, only relative cross section curves have been obtained for the O$^-$ and SO$^-$ ions. At higher energies, the ion pair thresholds are evaluated using the Wannier threshold law, yielding values consistent with those derived from thermochemical data.

physics.atm-clus↗

arXiv2025

TEDDY: A Family Of Foundation Models For Understanding Single Cell Biology

Understanding the biological mechanisms of disease is crucial for medicine, and in particular, for drug discovery. AI-powered analysis of genome-scale biological data holds great potential in this regard. The increasing availability of single-cell RNA sequencing data has enabled the development of large foundation models for disease biology. However, existing foundation models only modestly improve over task-specific models in downstream applications. Here, we explored two avenues for improving single-cell foundation models. First, we scaled the pre-training data to a diverse collection of 116 million cells, which is larger than those used by previous models. Second, we leveraged the availability of large-scale biological annotations as a form of supervision during pre-training. We trained the \model family of models comprising six transformer-based state-of-the-art single-cell foundation models with 70 million, 160 million, and 400 million parameters. We vetted our models on several downstream evaluation tasks, including identifying the underlying disease state of held-out donors not seen during training, distinguishing between diseased and healthy cells for disease conditions and donors not seen during training, and probing the learned representations for known biology. Our models showed substantial improvement over existing works, and scaling experiments showed that performance improved predictably with both data volume and parameter count.

cs.LG↗

arXiv2024

Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage?

This paper questions the effectiveness of a modern predictive uncertainty quantification approach, called \emph{evidential deep learning} (EDL), in which a single neural network model is trained to learn a meta distribution over the predictive distribution by minimizing a specific objective function. Despite their perceived strong empirical performance on downstream tasks, a line of recent studies by Bengs et al. identify limitations of the existing methods to conclude their learned epistemic uncertainties are unreliable, e.g., in that they are non-vanishing even with infinite data. Building on and sharpening such analysis, we 1) provide a sharper understanding of the asymptotic behavior of a wide class of EDL methods by unifying various objective functions; 2) reveal that the EDL methods can be better interpreted as an out-of-distribution detection algorithm based on energy-based-models; and 3) conduct extensive ablation studies to better assess their empirical effectiveness with real-world datasets. Through all these analyses, we conclude that even when EDL methods are empirically effective on downstream tasks, this occurs despite their poor uncertainty quantification capabilities. Our investigation suggests that incorporating model uncertainty can help EDL methods faithfully quantify uncertainties and further improve performance on representative downstream tasks, albeit at the cost of additional computational complexity.

cs.LG↗

arXiv2024

Detectors for Safe and Reliable LLMs: Implementations, Uses, and Limitations

Large language models (LLMs) are susceptible to a variety of risks, from non-faithful output to biased and toxic generations. Due to several limiting factors surrounding LLMs (training cost, API access, data availability, etc.), it may not always be feasible to impose direct safety constraints on a deployed model. Therefore, an efficient and reliable alternative is required. To this end, we present our ongoing efforts to create and deploy a library of detectors: compact and easy-to-build classification models that provide labels for various harms. In addition to the detectors themselves, we discuss a wide range of uses for these detector models - from acting as guardrails to enabling effective AI governance. We also deep dive into inherent challenges in their development and discuss future work aimed at making the detectors more reliable and broadening their scope.

cs.LG↗

arXiv2024

Surface Termination and Band Alignment in 2D Heterostructures

Heterostructures are ubiquitous in many optoelectronic devices and as photocatalysts. One of the key features of a heterojunction is the proper band alignment between the two materials. Estimation of the correct relative band positions with density functional theory (DFT) based electronic structure calculations is often constrained by the accuracy and cost associated with the various DFT functionals. In this study, we introduce a novel computational approach that achieves band alignments closely matching experimental results with the widely used PBE functional. We specifically examine the well-documented MoO3/MoS2 system, a type-II heterojunction. In our setup, the MoS2 layers are kept as it is but for MoO3 the individual layers are chosen differently. These alternative layers have higher surface energy, and hence, the band edges are higher than the conventional layers. This shift in band edges of the alternative MoO3 layers changes the band alignment in MoO3/MoS2 heterojunction from type-III to the experimentally observed type-II character. We also extend this computational strategy to additional systems, demonstrating its versatility and effectiveness.

cond-mat.mes-hall↗

arXiv2024

Thermometer: Towards Universal Calibration for Large Language Models

We consider the issue of calibration in large language models (LLM). Recent studies have found that common interventions such as instruction tuning often result in poorly calibrated LLMs. Although calibration is well-explored in traditional applications, calibrating LLMs is uniquely challenging. These challenges stem as much from the severe computational requirements of LLMs as from their versatility, which allows them to be applied to diverse tasks. Addressing these challenges, we propose THERMOMETER, a calibration approach tailored to LLMs. THERMOMETER learns an auxiliary model, given data from multiple tasks, for calibrating a LLM. It is computationally efficient, preserves the accuracy of the LLM, and produces better-calibrated responses for new tasks. Extensive empirical evaluations across various benchmarks demonstrate the effectiveness of the proposed method.

cs.LG↗

arXiv2024

$\textit{Trans-LoRA}$: towards data-free Transferable Parameter Efficient Finetuning

Low-rank adapters (LoRA) and their variants are popular parameter-efficient fine-tuning (PEFT) techniques that closely match full model fine-tune performance while requiring only a small number of additional parameters. These additional LoRA parameters are specific to the base model being adapted. When the base model needs to be deprecated and replaced with a new one, all the associated LoRA modules need to be re-trained. Such re-training requires access to the data used to train the LoRA for the original base model. This is especially problematic for commercial cloud applications where the LoRA modules and the base models are hosted by service providers who may not be allowed to host proprietary client task data. To address this challenge, we propose $\textit{Trans-LoRA}$ -- a novel method for lossless, nearly data-free transfer of LoRAs across base models. Our approach relies on synthetic data to transfer LoRA modules. Using large language models, we design a synthetic data generator to approximate the data-generating process of the $\textit{observed}$ task data subset. Training on the resulting synthetic dataset transfers LoRA modules to new models. We show the effectiveness of our approach using both LLama and Gemma model families. Our approach achieves lossless (mostly improved) LoRA transfer between models within and across different base model families, and even between different PEFT methods, on a wide variety of tasks.

cs.LG↗