arXiv subjects

Venkat Srinivasan

Publications and source records attributed to Venkat Srinivasan.

18 recordsLinked to original sources

arXiv2026

Gyan: An Explainable Neuro-Symbolic Language Model

Transformer based pre-trained large language models have become ubiquitous. There is increasing evidence to suggest that even with large scale pre-training, these models do not capture complete compositional context and certainly not, the full human analogous context. Besides, by the very nature of the architecture, these models hallucinate, are difficult to maintain, are not easily interpretable and require enormous compute resources for training and inference. Here, we describe Gyan, an explainable language model based on a novel non-transformer architecture, without any of these limitations. Gyan achieves SOTA performance on 3 widely cited data sets and superior performance on two proprietary data sets. The novel architecture decouples the language model from knowledge acquisition and representation. The model draws on rhetorical structure theory, semantic role theory and knowledge-based computational linguistics. Gyan's meaning representation structure captures the complete compositional context and attempts to mimic humans by expanding the context to a 'world model'. AI model adoption critically depends on trust and transparency especially in mission critical use cases. Collectively, our results demonstrate that it is possible to create models which are trustable and reliable for mission critical tasks. We believe our work has tremendous potential for guiding the development of transparent and trusted architectures for language models.

cs.CL↗

arXiv2026

Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence

We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, images, and video. Nemotron 3 Nano Omni delivers consistent accuracy improvements over its predecessor, Nemotron Nano V2 VL, across all modalities, enabled by advances in architecture, training data and recipes. In particular, Nemotron 3 delivers leading results in real-world document understanding, long audio-video comprehension, and agentic computer use. Built on the highly efficient Nemotron 3 Nano 30B-A3B backbone, Nemotron 3 Nano Omni further incorporates innovative multimodal token-reduction techniques to deliver substantially lower inference latency and higher throughput than other models of similar size. We are releasing model checkpoints in BF16, FP8, and FP4 formats, along with portions of the training data and codebase to facilitate further research and development.

cs.LG↗

arXiv2026

PivotRL: High Accuracy Agentic Post-Training at Low Compute Cost

Post-training for long-horizon agentic tasks has a tension between compute efficiency and generalization. While supervised fine-tuning (SFT) is compute efficient, it often suffers from out-of-domain (OOD) degradation. Conversely, end-to-end reinforcement learning (E2E RL) preserves OOD capabilities, but incurs high compute costs due to many turns of on-policy rollout. We introduce PivotRL, a novel framework that operates on existing SFT trajectories to combine the compute efficiency of SFT with the OOD accuracy of E2E RL. PivotRL relies on two key mechanisms: first, it executes local, on-policy rollouts and filters for pivots: informative intermediate turns where sampled actions exhibit high variance in outcomes; second, it utilizes rewards for functional-equivalent actions rather than demanding strict string matching with the SFT data demonstration. We theoretically show that these mechanisms incentivize strong learning signals with high natural gradient norm, while maximally preserving policy probability ordering on actions unrelated to training tasks. In comparison to standard SFT on identical data, we demonstrate that PivotRL achieves +4.17% higher in-domain accuracy on average across four agentic domains, and +10.04% higher OOD accuracy in non-agentic tasks. Notably, on agentic coding tasks, PivotRL achieves competitive accuracy with E2E RL with 4x fewer rollout turns. PivotRL is adopted by NVIDIA's Nemotron-3-Super-120B-A12B, acting as the workhorse in production-scale agentic post-training.

cs.AI↗

arXiv2025

From Bits to Rounds: Parallel Decoding with Exploration for Diffusion Language Models

Diffusion Language Models (DLMs) have recently emerged as a strong alternative to autoregressive language models (LMs). DLMs offer comparable accuracy with faster inference speed via parallel decoding. However, standard DLM decoding strategies relying on high-confidence tokens encounter an inherent information-theoretic bottleneck that restricts decoding progress and ultimately slows generation. We demonstrate both theoretically and empirically that prioritizing high-confidence tokens is inherently inefficient. High-probability tokens carry negligible information and strictly relying on them limits the effective progress made in each decoding round. We prove that the number of decoding rounds must grow linearly with the sample's total information (negative log-likelihood) and inversely with the per-round information budget, establishing a bits-to-rounds principle. We also propose Explore-Then-Exploit (ETE), a training-free decoding strategy that maximizes information throughput and decoding efficiency. ETE combines cross-block decoding with targeted exploration of high-uncertainty tokens to reshape the conditional distribution and trigger cascades of confident predictions. Experiments verify our theoretical bounds and demonstrate that ETE consistently reduces the required number of decoding rounds compared to confidence-only baselines without compromising generation quality.

cs.LG↗

arXiv2025

Ionic Interdiffusion at Cathode-Solid-Electrolyte Interface: A Machine Learning-Assisted Multiscale Investigation and Mitigation Strategies

Future lithium-based batteries are expected to use solid electrolytes to achieve higher energy density and fast charge capabilities. The majority of solid electrolytes are thermodynamically unstable against layered oxide cathodes. Here, the stability of LiCoO2 (LCO) cathode with Li10GeP2S12 (LGPS) solid electrolyte is investigated using ab initio molecular dynamics (AIMD) and machine learning molecular dynamics (MLMD). The propensity of ionic interdiffusion, formation of a passivation layer, and corresponding decay in cell performance is addressed using a continuum model. The large-scale MLMD simulations confirm that the LCO|LGPS interface permits interdiffusion of Co and other ionic species, leading to the formation and growth of a resistive interphase and dramatic capacity fade even in the first cycle. We then examine the literature evidence that incorporating a thin layer of LiNb0.5Ta0.5O3 (LNTO) between LCO and LGPS prevents the interdiffusion of ions. Atomistic simulations suggest that the substitution of Li in LNTO with Co is not thermodynamically favorable, which helps to minimize the ionic interdiffusion process. The stable Nb/Ta5+ states form a rigid metal-oxide framework, which consequently also prevents the substitution of Nb/Ta. However, continuum level analysis suggests that due to the higher mechanical stiffness of LNTO, interfacial delamination between the LCO and LNTO is possible, which can minimize the effectiveness of the protective layer. This paper suggests the need for the development of novel interlayers that balance low interdiffusion with low stiffness.

cond-mat.mtrl-sci↗

arXiv2025

Depth-resolved measurement of solvation entropy, interfacial transport and charge-transfer kinetics of practical lithium-ion batteries

Understanding the performance of electrochemical energy storage systems requires probing the electrochemical properties at each layer and interface during cell operation. While traditional onboard and operando methods can measure impedance, voltage, or capacity, they lack spatial resolution to pinpoint the properties to specific layers and interfaces. In this work, we describe an approach of using thermal waves to measure entropy change, transport resistance, and charge-transfer resistance with depth resolution of a few microns within an electrochemical cell. We achieve this by relating heat generation at multiple harmonics of an AC current to electrochemical processes and leveraging frequency dependence of thermal penetration depth for spatial resolution. We name this frequency domain spectroscopy of the thermal signatures of the electrochemical processes measured at multiple harmonics of the alternating current as Multi-harmonic ElectroThermal Spectroscopy (METS). This technique enables isolation and measurement of solvation entropy at individual electrode-electrolyte interfaces from the first harmonic (1ω) thermal signature and resolution of the overall interfacial impedance into charge-transfer and interface transport resistance components from the second harmonic (2ω) thermal signature. From this, we also demonstrate an operando measurement of the growth of the solid-electrolyte interphase (SEI) layer at the lithium-electrolyte interface and show that two chemically similar electrodes can have significantly different interfacial transport resistance based on the preparation of the electrodes. Additionally, the method is not specific to lithium-ion chemistry and can therefore be generalized for all electrochemical systems of interest.

physics.chem-ph↗

arXiv2025

On the Performance of an Explainable Language Model on PubMedQA

Large language models (LLMs) have shown significant abilities in retrieving medical knowledge, reasoning over it and answering medical questions comparably to physicians. However, these models are not interpretable, hallucinate, are difficult to maintain and require enormous compute resources for training and inference. In this paper, we report results from Gyan, an explainable language model based on an alternative architecture, on the PubmedQA data set. The Gyan LLM is a compositional language model and the model is decoupled from knowledge. Gyan is trustable, transparent, does not hallucinate and does not require significant training or compute resources. Gyan is easily transferable across domains. Gyan-4.3 achieves SOTA results on PubmedQA with 87.1% accuracy compared to 82% by MedPrompt based on GPT-4 and 81.8% by Med-PaLM 2 (Google and DeepMind). We will be reporting results for other medical data sets - MedQA, MedMCQA, MMLU - Medicine in the future.

cs.CL↗

arXiv2023

Training Large Language Models Efficiently with Sparsity and Dataflow

Large foundation language models have shown their versatility in being able to be adapted to perform a wide variety of downstream tasks, such as text generation, sentiment analysis, semantic search etc. However, training such large foundational models is a non-trivial exercise that requires a significant amount of compute power and expertise from machine learning and systems experts. As models get larger, these demands are only increasing. Sparsity is a promising technique to relieve the compute requirements for training. However, sparsity introduces new challenges in training the sparse model to the same quality as the dense counterparts. Furthermore, sparsity drops the operation intensity and introduces irregular memory access patterns that makes it challenging to efficiently utilize compute resources. This paper demonstrates an end-to-end training flow on a large language model - 13 billion GPT - using sparsity and dataflow. The dataflow execution model and architecture enables efficient on-chip irregular memory accesses as well as native kernel fusion and pipelined parallelism that helps recover device utilization. We show that we can successfully train GPT 13B to the same quality as the dense GPT 13B model, while achieving an end-end speedup of 4.5x over dense A100 baseline.

cs.LG↗

arXiv2023

Using thermal interface resistance for non-invasive operando mapping of buried interfacial lithium morphology in-solid state batteries

The lithium metal-solid-state electrolyte interface plays a critical role in the performance of solid-state batteries. However, operando characterization of the buried interface morphology in solid-state cells is particularly difficult because of the lack of direct optical access. Destructive techniques that require isolating the interface inadvertently modify the interface and cannot be used for operando monitoring. In this work, we introduce the concept of thermal wave sensing using modified 3-omega sensors that is attached to the outside of the lithium metal-solid state cells to non-invasively probe the morphology of the lithium metal-electrolyte interface. We show that the thermal interface resistance measured by the 3-omega sensors relates directly to the physical morphology of the interface and demonstrate that 3-omega thermal wave sensing can be used for non-invasive operando monitoring the morphology evolution of the lithium metal-solid state electrolyte interface.

physics.chem-ph↗

arXiv2017

Improving Part-of-Speech Tagging for NLP Pipelines

This paper outlines the results of sentence level linguistics based rules for improving part-of-speech tagging. It is well known that the performance of complex NLP systems is negatively affected if one of the preliminary stages is less than perfect. Errors in the initial stages in the pipeline have a snowballing effect on the pipeline's end performance. We have created a set of linguistics based rules at the sentence level which adjust part-of-speech tags from state-of-the-art taggers. Comparison with state-of-the-art taggers on widely used benchmarks demonstrate significant improvements in tagging accuracy and consequently in the quality and accuracy of NLP systems.

cs.CL↗

arXiv2013

On plastic deformation and fracture in Si films during electrochemical lithiation/delithiation cycling

An in situ study of deformation, fracture, and fatigue behavior of silicon as a lithium-ion battery electrode material is presented. Thin films (100-200 nm) of silicon are cycled in a half-cell configuration with lithium metal foil as counter/reference electrode, with 1M lithium hexafluorophosphate in ethylene carbonate, diethylene carbonate, dimethyl carbonate solution (1:1:1, wt.%) as electrolyte. Stress evolution in the Si thin-film electrodes during electrochemical lithiation and delithiation is measured by monitoring the substrate curvature using the multi-beam optical sensing method. The stress measurements have been corrected for contributions from residual stress arising from sputter-deposition. An indirect method for estimating the potential errors due to formation of the solid-electrolyte-interphase layer and surface charge on the stress measurements was presented. The films undergo extensive inelastic deformation during lithiation and delithiation. The peak compressive stress during lithiation was 1.48 GPa. The stress data along with the electron microscopy observations are used to estimate an upper bound fracture resistance of lithiated Si, which is approximately 9-11 J/m^2. Fracture initiation and crack density evolution as a function of cycle number is also reported.

cond-mat.mtrl-sci↗

arXiv2012

Analysis of Lithiation and Delithiation Kinetics in Silicon

Analysis of lithiation and delithiation kinetics in pulse-laser-deposited crystalline thin-film silicon (Si) electrodes is presented. Data from open-circuit relaxation experiments are used in conjunction with a model based on Tafel kinetics and double-layer capacitance to estimate the apparent transfer coefficients (αa, αc), and exchange current density to capacitance ratio (i0/Cdl) for lithiation and delithiation reactions in a lithiated silicon (LixSi) system. Parameters estimated from data sets obtained during first-cycle amorphization of crystalline Si, as well as from cycled crystalline Si and amorphous Si thin-film electrodes do not show much variation, indicating that they are intrinsic to lithiation/delithiation in Si. A methodology to estimate the side-reaction rate and its role in the evolution of the open-circuit potential of the LixSi system are discussed. We conclude that the large potential offset between lithiation and delithiation reactions at any given state of charge is partially caused by a large kinetic resistance (i.e., small i0). Using the estimated parameters, the model is shown to predict successfully the behavior of the system under galvanostatic lithiation and delithiation.

cond-mat.mtrl-sci↗

arXiv2011

Surface Structural Disordering in Graphite upon Lithium Intercalation/Deintercalation

We report on the origin of the surface structural disordering in graphite anodes induced by lithium intercalation and deintercalation processes. Average Raman spectra of graphitic anodes reveal that cycling at potentials that correspond to low lithium concentrations in LixC (0 \leq x < 0.16) is responsible for most of the structural damage observed at the graphite surface. The extent of surface structural disorder in graphite is significantly reduced for the anodes that were cycled at potentials where stage-1 and stage-2 compounds (x > 0.33) are present. Electrochemical impedance spectra show larger interfacial impedance for the electrodes that were fully delithiated during cycling as compared to electrodes that were cycled at lower potentials (U < 0.15 V vs. Li/Li+). Steep Li+ surface-bulk concentration gradients at the surface of graphite during early stages of intercalation processes, and the inherent increase of the LixC d-spacing tend to induce local stresses at the edges of graphene layers, and lead to the breakage of C-C bonds. The exposed graphite edge sites react with the electrolyte to (re)form the SEI layer, which leads to gradual degradation of the graphite anode, and causes reversible capacity loss in a lithium-ion battery.

cond-mat.mtrl-sci↗

arXiv2011

In Situ Measurements of Stress Evolution in Silicon Thin Films During Electrochemical Lithiation and Delithiation

We report in situ measurements of stress evolution in a silicon thin-film electrode during electrochemical lithiation and delithiation by using the Multi-beam Optical Sensor (MOS) technique. Upon lithiation, due to substrate constraint, the silicon electrode initially undergoes elastic deformation, resulting in rapid rise of compressive stress. The electrode begins to deform plastically at a compressive stress of ca. -1.75 GPa; subsequent lithiation results in continued plastic strain, dissipating mechanical energy. Upon delithiation, the electrode first undergoes elastic straining in the opposite direction, leading to a tensile stress of ca. 1 GPa; subsequently, it deforms plastically during the rest of delithiation. The plastic flow stress evolves continuously with lithium concentration. Thus, mechanical energy is dissipated in plastic deformation during both lithiation and delithiation, and it can be calculated from the stress measurements; we show that it is comparable to the polarization loss. Upon current interrupt, both the film stress and the electrode potential relax with similar time-constants, suggesting that stress contributes significantly to the chemical potential of lithiated-silicon.

cond-mat.mtrl-sci↗

arXiv2011

Lithium Diffusion in Graphitic Carbon

Graphitic carbon is currently considered the state-of-the-art material for the negative electrode in lithium-ion cells, mainly due to its high reversibility and low operating potential. However, carbon anodes exhibit mediocre charge/discharge rate performance, which contributes to severe transport-induced surface-structural damage upon prolonged cycling, and limits the lifetime of the cell. Lithium bulk diffusion in graphitic carbon is not yet completely understood, partly due to the complexity of measuring bulk transport properties in finite-sized, non-isotropic particles. To solve this problem for graphite, we use the Devanathan-Stachurski electrochemical methodology combined with ab-initio computations to deconvolute, and quantify the mechanism of lithium-ion diffusion in highly oriented pyrolytic graphite (HOPG). The results reveal inherent high lithium-ion diffusivity in the direction parallel to the graphene plane (ca. 10^-7 - 10^-6 cm2 s-1), as compared to sluggish lithium-ion transport along grain boundaries (ca. 10^-11 cm^2 s^-1), indicating the possibility of rational design of carbonaceous materials and composite electrodes with very high rate capability.

cond-mat.mtrl-sci↗

arXiv2011

Increased Cycling Efficiency and Rate Capability of Copper-coated Silicon Anodes in Lithium-ion Batteries

Cycling efficiency and rate capability of porous copper-coated, amorphous silicon thin-film negative electrodes are compared to equivalent silicon thin-film electrodes in lithium-ion batteries. The presence of a copper layer coated on the active material plays a beneficial role in increasing the cycling efficiency and the rate capability of silicon thin-film electrodes. Between 3C and C/8 discharge rates, the available cell energy decreased by 8% and 18% for 40 nm copper-coated silicon and equivalent silicon thin-film electrodes, respectively. Copper-coated silicon thin-film electrodes also show higher cycling efficiency, resulting in lower capacity fade, than equivalent silicon thin-film electrodes. We believe that copper appears to act as a glue that binds the electrode together and prevents the electronic isolation of silicon particles, thereby decreasing capacity loss. Rate capability decreases significantly at higher copper-coating thicknesses as the silicon active-material is not accessed, suggesting that the thickness and porosity of the copper coating need to be optimized for enhanced capacity retention and rate capability in this system.

cond-mat.mtrl-sci↗

arXiv2011

In Situ Measurements of Stress-Potential Coupling in Lithiated Silicon

An analysis of the dependence of electric potential on the state of stress of a lithiated-silicon electrode is presented. Based on the Larché and Cahn chemical potential for a solid solution, a thermodynamic argument is made for the existence of the stress-potential coupling in lithiated-silicon; based on the known properties of the material, the magnitude of the coupling is estimated to be ca. 60 mV/GPa in thin-film geometry. An experimental investigation is carried out on silicon thin-film electrodes in which the stress is measured in situ during electrochemical lithiation and delithiation. By progressively varying the stress through incremental delithiation, the relation between stress change and electric-potential change is measured to be 100 - 120 mV/GPa, which is of the same order of magnitude as the prediction of the analysis. The importance of the coupling is discussed in interpreting the hysteresis observed in potential vs. state-of-charge plots, and the role of stress in modifying the maximum charge capacity of a silicon electrode under stress.

cond-mat.mtrl-sci↗

arXiv2011

Real-time Measurement of Stress and Damage Evolution During Initial Lithiation of Crystalline Silicon

Crystalline to amorphous phase transformation during initial lithiation in (100) silicon-wafers is studied in an electrochemical cell with lithium metal as the counter and reference electrode. It is demonstrated that severe stress jumps across the phase boundary lead to fracture and damage, which is an essential consideration in designing silicon based anodes for lithium ion batteries. During initial lithiation, a moving phase boundary advances into the wafer starting from the surface facing the lithium electrode, transforming crystalline silicon into amorphous LixSi. The resulting biaxial compressive stress in the amorphous layer is measured in situ and it was observed to be ca. 0.5 GPa. HRTEM images reveal that the crystalline-amorphous phase boundary is very sharp, with a thickness of ~ 1 nm. Upon delithiation, the stress rapidly reverses, becomes tensile and the amorphous layer begins to deform plastically at around 0.5 GPa. With continued delithiation, the yield stress increases in magnitude, culminating in sudden fracture of the amorphous layer into micro-fragments and the cracks extend into the underlying crystalline silicon.

cond-mat.mtrl-sci↗