Publications

We report on the structural properties and optical and electrical characteristics of InN epitaxial layers grown on highly resistive GaN templates using migration enhanced metalorganic chemical vapor deposition (MEMOCVD). The material quality of InN improved significantly for the layer thickness larger than 150 nm. The highest extracted value of the room temperature electron mobility was close to 850 cm2/V s for samples with electron carrier concentration of ∼4×1018 cm−3. The measured dependence of the electron mobility on electron concentration is interpreted using the model accounting for ionized impurity scattering, polar optical scattering, and compensation. The MEMOCVD-grown material exhibited stronger photoluminescence (PL) compared to InN deposited using conventional metalorganic chemical vapor deposition. Room temperature PL spectra were similar to InN grown using molecular beam epitaxy (MBE) with peak emission at 0.8 eV. The obtained results demonstrate the potential of the MEMOCVD technique for deposition of high quality InN epitaxial layers at reduced temperatures comparable to those used in MBE growth.

AI agents increasingly excel at generating, testing, and refining code. However, they fall short on tasks requiring formal guarantees of full coverage that testing alone cannot provide. Distributed systems are a prime example: properties such as consistency between reads and writes must hold under every possible interleaving of events. Mechanized formal verification can guarantee such correctness, but typically demands months to years of expert effort. As evidence, even SOTA coding agents (Codex with GPT-5.4 and Claude Code with Opus 4.6) succeed on only 2/7 distributed key-value-store specifications. In this paper, we present the first effective approach to addressing this gap, Inductive Deductive Synthesis (IDS), which jointly and incrementally synthesizes implementation and proof, and learns from failed attempts to systematically try promising strategies. Built as an agentic LLM system, IDS achieves 7/7 in about 6.8 hours and $106 per spec on average, roughly 200x faster than expert effort and 17% cheaper than SOTA agents. IDS further incorporates performance feedback into the same loop, yielding implementations up to 3x faster than published verified systems.

Sequential pulling policies to flatten and smooth fabrics have applications from surgery to manufacturing to home tasks such as bed making and folding clothes. Due to the complexity of fabric states and dynamics, we apply deep imitation learning to learn policies that, given color (RGB), depth (D), or combined color-depth (RGBD) images of a rectangular fabric sample, estimate pick points and pull vectors to spread the fabric to maximize coverage. To generate data, we develop a fabric simulator and an algorithmic supervisor that has access to complete state information. We train policies in simulation using domain randomization and dataset aggregation (DAgger) on three tiers of difficulty in the initial randomized configuration. We present results comparing five baseline policies to learned policies and report systematic comparisons of RGB vs D vs RGBD images as inputs. In simulation, learned policies achieve comparable or superior performance to analytic baselines. In 180 physical experiments with the da Vinci Research Kit (dVRK) surgical robot, RGBD policies trained in simulation attain coverage of 83% to 95% depending on difficulty tier, suggesting that effective fabric smoothing policies can be learned from an algorithmic supervisor and that depth sensing is a valuable addition to color alone. Supplementary material is available at https://sites.google.com/view/fabric-smoothing.

Infrequent stimulation of a subject generates event-related potential (ERP) signals masked by background EEG activity. It is generally assumed that this background activity is normal white noise. Another assumption made is that the underlying evoked signals are deterministic and they do not vary from trial to trial. The signal, thus, is modeled as y/sub i//sup (j)/=x/sub i/+/spl sigma//sub i//sup (j)/, where y/sub i//sup (j)/ is the jth trial and z/sub i/ is a white noise. The signal x/sub i/ is recovered by averaging the observations y/sub i//sup (j)/. In reality, the background activity is "colored" and not always Gaussian. The time samples of the background activity are generally correlated. In addition, the signal x/sub i/ varies across observations. The purpose of this study is to extract single trial ERPs from the EEG. The authors are interested in the trial to trial variation of the ERPs and their clinical applicability. They have investigated the wavelet soft-thresholding method to remove the background noise and separate out the single trial response. The non-white background activity, after wavelet transformation, may concentrate in certain resolution levels. The authors determined these levels by testing the Gaussianity of the wavelet coefficients, using both the /spl chi//sup 2/ and Kolmogorov-Smirnov goodness of fit tests. In the resolution levels at which the null hypothesis was not rejected, the noise level was estimated. The de-noising threshold was then calculated using a level dependent rule T/sub j,N/=/spl radic/(2 log(N))/spl middot/MAD(C/sub j,k/)/0.6745, where C/sub j,k/ are the wavelet coefficients and MAD(C/sub j,k/)=Median(|C/sub j,k|/) is an estimator of the noise level. The resolution levels at which the null hypothesis was rejected were not thresholded.

Actions as simple as grasping an object or navigating around it require a rich understanding of that object's 3D shape from a given viewpoint. In this paper we repurpose powerful learning machinery, originally developed for object classification, to discover image cues relevant for recovering the 3D shape of potentially unfamiliar objects. We cast the problem as one of local prediction of surface normals and global detection of 3D reflection symmetry planes, which open the door for extrapolating occluded surfaces from visible ones. We demonstrate that our method is able to recover accurate 3D shape information for classes of objects it was not trained on, in both synthetic and real images.

High stiffness and strength carbon fibres are commonly used to reinforce epoxy-resin composites. While wild Antheraea pernyi silk fibres exhibit high toughness originating from their α-helix/random coil conformation structures and their micro-fibre morphology, their insufficient strength and stiffness hinders them from being used in similar structural composites. In this work, we use interply hybridization of silk and carbon fibres to reinforce epoxy-matrix composites. With increased carbon fibre content, the quasi-static tensile/flexural stiffness and strength increases following the rule of mixtures while more silk fibre acts to increase ductility and impact strength. This results in a composite comprising equal volumes of carbon and silk fibres achieving an impact strength of 98 kJ m −2 , which is twice that of purely carbon-fibre reinforced composites (44 kJ m −2 ). This work shows tough natural silk fibres and strong synthetic fibres can be successfully integrated into epoxy-resin composites for tailored mechanical properties.

The internal states of most deep neural networks are difficult to interpret, which makes diagnosis and debugging during training challenging. Activation maximization methods are widely used, but lead to multiple optima and are hard to interpret (appear noise-like) for complex neurons. Image-based methods use maximally-activating image regions which are easier to interpret, but do not provide pixel-level insight into why the neuron responds to them. In this work we introduce an MCMC method: Langevin Dynamics Activation Maximization (LDAM), which is designed for diagnostic visualization. LDAM provides two affordances in combination: the ability to explore the set of maximally activating pre-images, and the ability to trade-off interpretability and pixel-level accuracy using a GAN-style discriminator as a regularizer. We present case studies on MNIST, CIFAR and ImageNet datasets exploring these trade-offs. Finally we show that diagnostic visualization using LDAM leads to a novel insight into the parameter averaging method for deep net training.