A5037987827- X-ray Diffraction in Crystallography
- Crystallization and Solubility Studies
- Crystallography and molecular interactions
- Electronic and Structural Properties of Oxides
- Ferroelectric and Piezoelectric Materials
- Multiferroics and related materials
- Semiconductor materials and devices
- Advancements in Battery Materials
- Topic Modeling
- Electron and X-Ray Spectroscopy Techniques
- Advanced X-ray Imaging Techniques
- Advanced Condensed Matter Physics
- Natural Language Processing Techniques
- Photorefractive and Nonlinear Optics
- Magnetic and transport properties of perovskites and related materials
- Advanced Electron Microscopy Techniques and Applications
- Magnetic properties of thin films
- Spectroscopy and Quantum Chemical Studies
- Ion-surface interactions and analysis
- Advanced Battery Materials and Technologies
- Machine Learning in Materials Science
- ZnO doping and properties
- Metal-Organic Frameworks: Synthesis and Applications
- Geophysics and Sensor Technology
- Ammonia Synthesis and Nitrogen Reduction
University of Colorado Boulder
2024-2025
University of Colorado System
2025
University of Wisconsin–Madison
2019-2024
Oxford Instruments (United Kingdom)
2020
Madison Group (United States)
2020
Argonne National Laboratory
2019
Reinforcement learning from human feedback (RLHF) is a technique for training AI systems to align with goals. RLHF has emerged as the central method used finetune state-of-the-art large language models (LLMs). Despite this popularity, there been relatively little public work systematizing its flaws. In paper, we (1) survey open problems and fundamental limitations of related methods; (2) overview techniques understand, improve, complement in practice; (3) propose auditing disclosure...
Hard carbon (HC) is a leading anode material for sodium-ion batteries, but its complex microstructure complicates understanding of sodium storage mechanisms. Using X-ray total scattering and density functional theory calculations, this study clarifies how HC’s microstructural variations influence across the slope (high potential) plateau (low regions potential-capacity curve. In region, initially adsorbs at high-binding energy defect sites subsequently intercalates between graphene layers,...
Abstract Understanding the formation and evolution of cathode‐electrolyte interphase (CEI), which forms at interface between cathode electrolyte, is crucial for revealing degradation mechanisms in materials, especially developing strategies to stabilize strongly oxidizing conditions that evolve high operating voltages next‐generation Li‐ion batteries. However, The present understanding CEI challenged by its complex dynamic nature. In this work, near‐edge X‐ray absorption fine structure...
Abstract Domain wall structures form spontaneously due to epitaxial misfit during thin film growth. Imaging the dynamics of domains and domain walls at ultrafast timescales can provide fundamental clues features that impact electrical transport in electronic devices. Recently, deep learning based methods showed promising phase retrieval (PR) performance, allowing intensity-only measurements be transformed into snapshot real space images. While Fourier imaging model involves complex-valued...
Abstract Material functionality can be strongly determined by structure extending only over nanoscale distances. The pair distribution function presents an opportunity for structural studies beyond idealized crystal models and to investigate varying length scales. Applying this method with ultrafast time resolution has the potential similarly disrupt study of dynamics phase transitions. Here we demonstrate such a measurement CuIr 2 S 4 optically pumped from its low-temperature Ir-dimerized...
Epitaxial crystallization of complex oxides provides the means to create materials with precisely selected composition, strain, and orientation, thereby controlling their functionalities. Extending this control nanoscale three-dimensional geometries can be accomplished via a analog oxide solid-phase epitaxy, lateral epitaxial crystallization. The orientation crystals within laterally crystallized SrTiO
Above-bandgap optical illumination induces a transformation from tilted tetragonal-like (T-like) and rhombohedral-like (R-like) phases to an untilted T-like phase in compressively strained BiFeO3. Optical excitation leads out-of-plane lattice expansion the phase. The proceeds regions with boundaries between R-like phases, consistent motion of boundary. optically induced indicates that there are new driven routes towards nanosecond-scale control transformations ferroelectrics multiferroics.
Schematic depiction of the CO<sub>2</sub> adsorption-induced lattice distortion and ligand rotation in CPL-5.
We introduce methods for discovering and applying sparse feature circuits. These are causally implicated subnetworks of human-interpretable features explaining language model behaviors. Circuits identified in prior work consist polysemantic difficult-to-interpret units like attention heads or neurons, rendering them unsuitable many downstream applications. In contrast, circuits enable detailed understanding unanticipated mechanisms. Because they based on fine-grained units, useful tasks:...
$\mathrm{SrV}{\mathrm{O}}_{3}$ thin films with a high figure of merit for applications as transparent conductors were crystallized from amorphous layers using solid phase epitaxy (SPE). Epitaxial on $\mathrm{SrTi}{\mathrm{O}}_{3}$ SPE exhibit room-temperature resistivities low $5.2\ifmmode\times\else\texttimes\fi{}{10}^{--5}$ and $2.5\ifmmode\times\else\texttimes\fi{}{10}^{--5}\phantom{\rule{0.16em}{0ex}}\mathrm{\ensuremath{\Omega}}\phantom{\rule{0.16em}{0ex}}\mathrm{cm}$, residual...
Weakly coupled ferroelectric/dielectric superlattice thin-film heterostructures exhibit complex nanoscale polarization configurations that arise from a balance of competing electrostatic, elastic, and domain-wall contributions to the free energy. A key feature these is can locally have significant component along surface normal direction with an overall configuration maintaining zero net in-plane polarization. PbTiO3/SrTiO3 on conducting SrRuO3 bottom electrode SrTiO3 room-temperature stripe...
A pillared layer network containing amide functional groups (Cu(pzdc)(pia); pzdc = pyrazine-2,3-dicarboxylate; pia N-(4-pyridyl)isonicotinamide) was used to test a postsynthesis metalation rationale insert lithium and create porous surface with enhanced CO2 adsorption capacity. Synchrotron powder X-ray diffraction (XRD) determine variations after lithiation in long-range textural properties. measurements at room temperature showed concave up isotherm shape an increasing high pressures,...
Optical excitation leads to ultrafast stress generation in the prototypical multiferroic BiFeO3. The time scales of are set by dynamics population excited electronic states and coupling configuration structure. X-ray free-electron laser diffraction reveals high-wavevector subpicosecond-time scale following ultraviolet a BiFeO3 thin film. Stress includes fast component with 1/e rise an upper limit 300 fs longer-rise components extending 1.5 ps. contributions delayed vary as function optical...
ScAlMgO4 (SAM) is a promising substrate material for group III-nitride semiconductors. SAM has lower lattice mismatch with materials compared to conventionally used sapphire (Al2O3) and silicon substrates. Bulk the issues of high cost lack large area Utilizing solid-phase epitaxy transform an amorphous on into crystalline form cost-efficient scalable approach. Amorphous layers were deposited 0001-oriented Al2O3 by sputtering crystallized annealing at temperature greater than 850 °C....
Intense hard x-ray pulses from a free-electron laser induce irreversible structural damage in perovskite oxide epitaxial heterostructure when pulse fluences exceed threshold value. The intensity of diffraction 25-nm thick BiFeO3 layer on SrTiO3 (STO) substrate measured using series decreases abruptly with per-pulse fluence 2.7 × 106 photons μm−2 at photon energy 9.7 keV but remains constant for 1.3 or less. resulted the destruction thin film within focal spot area and formation deep cavity...
The crystallization of amorphous complex oxide layers from isolated seed crystals presents an opportunity to remove geometric constraints posed by thin-film epitaxial growth methods employing single-crystal substrates. processes initiated a distribution nanoscale seeds occur in state mechanical stress that is different planar epitaxy. effects this were probed the model perovskite SrTiO3 nucleated seeds. Synchrotron nanobeam scattering and diffraction used probe spatial crystalline SrTiO3....
Concept erasure in language models has traditionally lacked a comprehensive evaluation framework, leading to incomplete assessments of effectiveness methods. We propose an paradigm centered on three critical criteria: innocence (complete knowledge removal), seamlessness (maintaining conditional fluent generation), and specificity (preserving unrelated task performance). Our metrics naturally motivate the development Erasure Language Memory (ELM), new method designed address all dimensions....
Lithium-ion batteries (LIBs) have been widely employed as energy storage devices in portable electronics and electric vehicles. Many processes occurring at the electrode/electrolyte interphases lead to performance degradation over time yet remain poorly understood. We demonstrate new methods based on scanning electrochemical microscopy (SECM) characterize LIB electrolyte oxidation, which is a key process cathode/electrolyte interphase. Our technique leverages combination of feedback mode...
All-solid-state batteries (ASSBs) are poised to meet the rising demand for high energy density electrochemical storage and safety in lithium-ion battery technology, a crucial aspect widespread adoption of electric vehicles [1]. However, solid-electrolyte interphase (SEI) layer arising from chemical incompatibility narrow solid electrolyte window poses significant hurdle. The SEI contributes resistance ion transfer resulting poor performance cyclability promising solid-state materials...
We introduce NNsight and NDIF, technologies that work in tandem to enable scientific study of very large neural networks. is an open-source system extends PyTorch deferred remote execution. NDIF a scalable inference service executes requests, allowing users share GPU resources pretrained models. These are enabled by the intervention graph, architecture developed decouple experiment design from model runtime. Together, this framework provides transparent efficient access internals deep...
What latent features are encoded in language model (LM) representations? Recent work on training sparse autoencoders (SAEs) to disentangle interpretable LM representations has shown significant promise. However, evaluating the quality of these SAEs is difficult because we lack a ground-truth collection that expect good recover. We thus propose measure progress dictionary learning by working setting LMs trained chess and Othello transcripts. These settings carry natural collections -- for...
Interpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity the field: most studies employ ad-hoc evaluations do not share theoretical foundations, making it difficult to measure progress compare pros cons of different techniques. Furthermore, while mechanistic frequently discussed, basic causal units underlying these mechanisms are often explicitly defined. In this paper, we propose perspective on interpretability...
Sparse Autoencoders (SAEs) are an interpretability technique aimed at decomposing neural network activations into interpretable units. However, a major bottleneck for SAE development has been the lack of high-quality performance metrics, with prior work largely relying on unsupervised proxies. In this work, we introduce family evaluations based SHIFT, downstream task from Marks et al. (Sparse Feature Circuits, 2024) in which spurious cues removed classifier by ablating features judged to be...