## Page Views on Nuit Blanche since July 2010

My papers on ArXiv:
Approximating Kernels at the speed of Light
&
Imaging with Nature

LightOn
LinkedIn (557)|| on CrunchBase || our Blog
(2403)
(3984)
(13335)||
Attendant references pages:
The Advanced Matrix Factorization Jungle Page ||

Paris Machine Learning
@Meetup.com (7800 members) || @Archives

## Tuesday, August 20, 2019

### Transfer Learning as a Tool for Reducing Simulation Bias: Application to Inertial Confinement Fusion

** Nuit Blanche is now on Twitter: @NuitBlog **

Using Transfer learning for exploration purposes in expensive Inertial Confinement Fusion experiments is probably the only way to speed up our exploration of the right parameters in this nuclear fusion quest.

Transfer Learning as a Tool for Reducing Simulation Bias: Application to Inertial Confinement Fusion. by B. Kustowski , Jim A. Gaffney , Brian K. Spears , Gemma J. Anderson , Jayaraman Jayaraman Thiagarajan , and Rushil Anirudh

We adapt a technique, known in the machine learning community as transfer learning, to reduce the bias of a computer simulation using very sparse experimental data. Unlike the Bayesian calibration, which is commonly used to estimate the simulation bias, transfer learning involves calculating an artificial neural network surrogate model of the simulations. Assuming that the simulation code correctly predicts trends in the experimental data but it is subject to unknown biases, we then partially retrain, or transfer learn, the initial surrogate model to match the experimental data. This process eliminates the bias while still taking advantage of the physics relations learned from the simulation. Transfer learning can be easily adapted to a wide range of problems in science and engineering. In this paper, we carry out numerical tests to investigate the applicability of this technique to predict inertial confinement fusion experiments under new conditions. Using our synthetic validation data set we demonstrate that an accurate predictive model can be built by retraining an initial surrogate model with experimental data volumes so small that they are relevant to the inertial confinement fusion problem. This opens up new opportunities for knowledge transfer and building predictive models in physics. After implementing transfer learning in a standard neural network, we successfully extended the method to a more complex, generative adversarial network architecture, which will be needed for predicting not only scalars but also diagnostic images in our future work.

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

## Monday, August 19, 2019

### Enhanced Seismic Imaging with Predictive Neural Networks for Geophysics

** Nuit Blanche is now on Twitter: @NuitBlog **

This week, we will look into how some inverse problems.

We propose a predictive neural network architecture that can be utilized to update reference velocity models as inputs to full waveform inversion. Deep learning models are explored to augment velocity model building workflows during 3D seismic volume reprocessing in salt-prone environments. Specifically, a neural network architecture, with 3D convolutional, de-convolutional layers, and 3D max-pooling, is designed to take standard amplitude 3D seismic volumes as an input. Enhanced data augmentations through generative adversarial networks and a weighted loss function enable the network to train with few sparsely annotated slices. Batch normalization is also applied for faster convergence. Moreover, a 3D probability cube for salt bodies is generated through ensembles of predictions from multiple models in order to reduce variance. Velocity models inferred from the proposed networks provide opportunities for FWI forward models to converge faster with an initial condition closer to the true model. In each iteration step, the probability cubes of salt bodies inferred from the proposed networks can be used as a regularization term in FWI forward modelling, which may result in an improved velocity model estimation while the output of seismic migration can be utilized as an input of the 3D neural network for subsequent iterations.

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

## Friday, August 16, 2019

### Job: Several postdocs, Ground Breaking Deep Learning Technology for Monitoring the Brain during Surgery with Commercialization Opportunity, University of Pittsburgh

** Nuit Blanche is now on Twitter: @NuitBlog **

Kayhan just sent me the following:

Dear Igor,

I hope you are doing well.

I don't know if you remember me but we have been in contact a few times while I was a Ph.D. student at UPenn.

My lab at the University of Pittsburgh has several postdoc positions open. More specifically, I would be thankful if you could advertise this position (Link: https://kayhan.dbmi.pitt.edu/sites/default/files/JobAd.pdf) to your audience.

Best,
Kayhan

Sure Kayhan, I remember! Here is the announcement:
Ground Breaking Deep Learning Technology for Monitoring the Brain during Surgery with Commercialization Opportunity

We are developing a clinical tool based on deep learning to automatically detect stroke during surgery and alert the surgical team to avert complications and save lives. We are uniquely positioned at the intersection of the largest health care system in the US, the University of Pittsburgh Medical Center (UPMC), and top ranked academic institutions, the University of Pittsburgh (Pitt) and the Carnegie Mellon University (CMU). Our group consists of Pitt and UPMC faculty members who have complementary expertise in machine learning and in healthcare and specifically in deep learning, clinical informatics, neurology, and surgery. We develop novel deep learning and other machine learning methods for application to challenging clinical problems. We are very well funded by NIH, NSF, industry, and internal institutional grants.
In the current project, we are developing a clinical tool that will automatically detect stroke and other adverse events during surgery from an array of monitoring information, and provide highly accurate real time alerts to the surgical team to make course corrections during surgery. The clinical tool is to be deployed in operating rooms for monitoring surgeries and providing high quality alerts.
The successful candidate will work with us in a highly collaborative environment that spans the computer laboratory and the operating room and will gain unique and valuable experience in deep learning, development of a tool for a clinical setting, and in commercialization.
Expected qualifications Genuinely motivated to develop and apply machine learning to clinical problems. Strong expertise in machine learning is required; expertise in statistics and experience with messy clinical data is a plus. Python fluency is required. Demonstrated ability to make meaningful contributions to projects with a research flavor is valuable.
Experience/Abilities
• Hands-on experience building predictive models
• Experience working with diverse data types including signal and structured data; experience with text data is a plus
• Experience in programming in Python; experience in additional languages (R, C/C++) is a plus
• Aware of current best practices in machine learning
• Fluency in one of the deep learning frameworks is a plus (PyTorch or Tensorflow)
• Knowledge of statistics, including hypothesis testing with parametric and non-parametric tests and basic probability
• PhD in computer science, electrical engineering, statistics or equivalent computational / quantitative fields (exceptional MS candidates will be considered)
The goal of this project is to develop, evaluate and commercialize a tool for automatic detection of stroke during surgery. The successful candidate will have the rare opportunity to perform cutting-edge deep learning research and participate in a commercial endeavor.
If interested, contact Shyam Visweswaran, MD, PhD at shv3@pitt.edu and Kayhan Batmanghelich, PhD at kayhan@pitt.edu. For details of ongoing research work, visit http://www.thevislab.com/ and https://kayhan.dbmi.pitt.edu/. The University of Pittsburgh is an Affirmative Action/Equal Opportunity Employer and values equality of opportunity, human dignity, and diversity.

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

### Jobs: PhD scholarship on Algorithms for Event-Driven Camera Analysis at Western Sydney University, Australia

** Nuit Blanche is now on Twitter: @NuitBlog **

Paul Hurley just let me know of the following PhD scholarship

Hi Igor -- I don't know if you still put jobs/PhD scholarships on nuit blanche, but if you still do, would you mind sharing mine? It's an opportunity to build up foundational work for event-based cameras. https://www.westernsydney.edu.au/graduate_research_school/graduate_research_school/scholarships/current_scholarships/current_scholarships/scem_algorithms_for_event-driven_camera_analysis
Sure Paul ! Here is how the ad starts:

SCEM: Algorithms for Event-Driven Camera Analysis
School of Computing, Engineering and Mathematics
Scholarship code: 2019-089
Event-driven cameras are exciting technology that do not acquire full images like traditional cameras, but record only intensity changes when they occur. The International Centre for Neuromorphic Systems at Western Sydney University has been adapting them to perform Neuromorphic space imaging.
This PhD scholarship builds on this work to help develop the correct abstraction and a theory so as to improve knowledge extraction algorithms. It goes from modelling to algorithm testing using real data, working together with a world-class team.

What does the scholarship provide?
• Domestic candidates will receive a tax-free stipend of $30,000(AUD) per annum for up to 3 years to support living costs, supported by the Research Training Program (RTP) Fee Offset. • International candidates will receive a tax-free stipend of$30,000(AUD) per annum for up to 3 years to support living costs. Those with a strong track record will be eligible for a tuition fee waiver.
• Support for conference attendance, fieldwork and additional costs as approved by the School.
International candidates are required to hold an Overseas Student Health Care (OSHC)(opens in new window)insurance policy for the duration their study in Australia. This cost is not covered by the scholarship.

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

## Thursday, August 15, 2019

### Jobs: 2 PhD and RA positions at University of Luxembourg

** Nuit Blanche is now on Twitter: @NuitBlog **

Kumar also sent me the following announcements for different positions:

Dear Igor,
I was wondering if you could post on Nuit-Blanche the announcement of the following Ph.D./R.A. positions at SnT, University of Luxembourg on signal processing for next-generation radar systems.
Thanks!
--
Regards,

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

### Hardware realization of a CS-based MIMO radar

** Nuit Blanche is now on Twitter: @NuitBlog **

Kumar just sent me the following the other day:

Hi Igor,
We recently published our work on the hardware realization of a CS-based MIMO radar in IEEE Transactions on Aerospace and Electronic Systems. Your readers might be interested in this.
https://ieeexplore.ieee.org/abstract/document/8743424
--
Regards,
Kumar Vijay Mishra
Thanks Kumar  !

Here is the abstract:

We present a cognitive prototype that demonstrates a colocated, frequency-division-multiplexed, multiple-input multiple-output (MIMO) radar which implements both temporal and spatial sub-Nyquist sampling. The signal is sampled and recovered via the Xampling framework. Cognition is due to the fact that the transmitter adapts its signal spectrum by emitting only those subbands that the receiver samples and processes. Real-time experiments demonstrate sub-Nyquist MIMO recovery of target scenes with 87:5% spatio-temporal bandwidth reduction and signal-to-noise-ratio of -10 dB.

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email.

## Friday, July 12, 2019

### Two new preprints using LightOn Optical Processing Unit (OPU)

** Nuit Blanche is now on Twitter: @NuitBlog **

Two new preprints using @LightOn's Optical Processing Unit. The first one shows a speed-up of 400 times faster than a CPU implementation. Both of these preprints used @LightOn's Optical Processing Unit. If you are interested in being part of our pilot program to have access to our Light Processing technology, please register at https://www.lighton.ai/lighton-cloud/ All papers and preprints utilizing our Light processing technology can be found at: https://www.lighton.ai/our-technology/

Reservoir Computing is a relatively recent computational framework based on a large Recurrent Neural Network with fixed weights. Many physical implementations of Reservoir Computing have been proposed to improve speed and energy efficiency. In this study, we report new advances in Optical Reservoir Computing using multiple light scattering to accelerate the recursive computation of the reservoir states. Two different spatial light modulation technologies, namely, phase or binary amplitude modulations, are compared. Phase modulation is a promising direction already employed in other photonic implementations of Reservoir Computing. Additionally, we report a Digital-Micromirror-based Reservoir Computing at up to 640 Hz, more than double the previously reported frequency using a remotely controlled optical device developed by LightOn, and present new binarization strategies to improve the performance of binarized Reservoir Computing.
and another one, on making our Optical Processing Unit linear.

In this paper we tackle the problem of recovering the phase of complex linear measurements when only magnitude information is available and we control the input. We are motivated by the recent development of dedicated optics-based hardware for rapid random projections which leverages the propagation of light in random media. A signal of interest ξRN is mixed by a random scattering medium to compute the projection y=Aξ, with ACM×N being a realization of a standard complex Gaussian iid random matrix. Two difficulties arise in this scheme: only the intensity |y|2 can be recorded by the camera, and the transmission matrix A is unknown. We show that even without knowing A, we can recover the unknown phase of y for some equivalent transmission matrix with the same distribution as A. Our method is based on two observations: first, changing the phase of any row of A does not change its distribution; and second, since we control the input we can interfere ξ with arbitrary reference signals. We show how to leverage these observations to cast the measurement phase retrieval problem as a Euclidean distance geometry problem. We demonstrate appealing properties of the proposed algorithm on both numerical simulations and in real hardware experiments. Not only does our algorithm accurately recover the missing phase, but it mitigates the effects of quantization and the sensitivity threshold, thus also improving the measured magnitudes.

all papers and preprints utilizing our Light processing technology can be found at: https://www.lighton.ai/our-technology/