A Machine Learning and Data Science Blog

A study of model aggregator algorithms in Federated Learning

noreply@blogger.com (psakamoori) — Mon, 27 Jun 2022 18:34:00 +0000

What is Federated learning (FL)?

Distributed model training in the edge. The principle of this approach is to aggregate models learned on distributed clients in order to obtain a new, more general model. The results model is then redistributed to clients for further training.

FL struggles with Data Heterogeneity? As each client train a model in federation with non-identical and individual (non-iid) data, during model aggregation the next version of aggregated model might have reduced accuracy when deployed in client settings due to non-iid.

Goals of Federated Learning Model aggregator? Heterogeneity of devices and user’s challenges ML with double objective • Generalization: Achieving over all target accuracy with federated learning settings (pervasive devices). • Personalization: Achieving target accuracy on each client in AI tasks with already seen data or new data.

Strategies adopted in Model Aggregated algorithms? Key point in federated learning is the way specialized models are aggregated at the server In case of deep learning – two families of algorithms implementing different strategies are considered • First Strategy: Emphasize generalization – aggregation algorithm considers local models and build a new model that potentially calls into question all layers and all weights associated with neurons. This approach is exemplified by FedAvg and FedMA • Second Strategy: In contrast focus more on client specialization. Thus, the algorithm does not question certain parts of the local models. Specifically, only the local models base layers are sent to the server for generalization, while the last layers are kept unchanged. This approach is exemplified by the FedPer algorithm.

Different type of model aggregator algorithms available/used?

Federated Averaging: (FedAvg) • Starts with random initialization of a neural model (server is in charge of coordinating and managing model transfers with client devices) • Resulting model is sent to clients to start local training from it • When on-device training is finished the weights of the local models are sent to the server • Aggregation is done in a weighted averaging manner where clients with more data influence more significantly the newly aggregated model. • FedAvg is however, has naive form of aggregation due to its coordinate wise averaging that may lead to sub-optimal solutions. Due to non-iiD data neurons in the same coordinate may be opted for entirely different purposes due to client’s specialization • Averaging neurons that are drastically different causes decrement results.

Federated learning with Personalization layers (FedPer): • Similar to FedAvg the way it computes new weights in the aggregated model. • However, it differs strongly on the parts of the model that are considered during aggregation • Clients only communicate the neural model’s base layer to the server and retrain the other layers • Underlying idea is base layer deals with “representation learning” and can be advantageously shared by clients through aggregation • Upper layers are more concerned with decision making which is more specific to each client • FedPer clients better handle various inputs (in the base layers) while being able to specialize in their particular data (in the upper layers). • However, aggregated model, on the server-side is only partial and is not usable for decision-making due to missing layers • FedPer can be seen as an adaption of the Transfer Learning methodology into a federared learning scheme. Studies have shown thtat it can surpass centralized learning and FedAvg approach in the HAR field.

Federated Matched Averaging (FedMA): • *FedMA modifies the neural model architecture by incorporating a layer-wise aggregation process where similar neurons can be fused, and new ones can be added. • This approach treats the number of nodes in a layer as a sub-problem to solve rather than a hyper-parameter to be set as an extension to CNN and RNN • *FedMA Considers that neuron in the NN layer are permutation invariant (changing the layer/neuron order will not impact the NN output). • *The algorithm central intuition is that all clients can contain neurons that are similar and should be merged together. **All neurons in the same cluster are averaged to produce a global neuron** • [[To find out which neuron can be fused, the algorithm uses a 2D permutation matrix that is computed iteratively from increasing rank layers.]] • Experiments with deep CNN and LSTM memory architectures show that the FedMA algorithm outperforms FedAvg on CV datasets.

Fededated Distance (FedDist): • Novel neuron matching and detail a federated learning algorithm based on a Euclidian distance dissimilarity measurement. • This algorithm includes some elements of FedAvg and FedMA, is called Federated Distance (FedDist) for its emphasis on computing distances of neurons of similar coordinates when comparing clients and server models. • FedDist recognized, like FedMA, that some client models may diverge because of heterogeneous, non-IID data. This results in neurons that cannot be matched with neurons from other models (because of weights that are too far apart). • FedDist also recognized that the model structure is relatively stable, which means that neurons with the same coordinates play a similar role. This view provides an opportunity to build a coordinate-wise approach • FedDist identifies diverging neurons using Euclidean distances. These neurons that are specific to certain clients are added to the aggregated model as new neurons. • This new neuron adding scheme can lead to larger models that are able to generalize better. As new neurons are added to a layer, a layer-wise training round is added in order to allow the neurons in the next layers to adjust to the new incoming neurons and weights. • To do that layer with the new neuron and those below are frozen, and the subsequent layers are trained. When all layers have been treated, and a new aggregated model is computed. ◦ A global server model is computed from all clients. ◦ Outliers are identified using the Euclidean distance and added to the aggregated model. • A penalty function has also been implemented to raise the threshold as training continues to prevent the never-ending addition of new neurons. If a neuron in any of the clients holds an individual distance above the threshold, it is then added to the server model. The process is performed layer wise. At each communication round, it is performed on the first layer.

Reference: https://arxiv.org/abs/2110.10223 Happy reading..

Tutorial: RealSense Python SDK on Ubuntu

noreply@blogger.com (psakamoori) — Mon, 01 Feb 2021 18:56:00 +0000

Container Orchestration using Kubernetes (K8S) with sample code

noreply@blogger.com (psakamoori) — Mon, 01 Feb 2021 18:05:00 +0000

How to recover completely broken Ubuntu 18.04

noreply@blogger.com (psakamoori) — Tue, 27 Oct 2020 04:59:00 +0000

My upgrade from 16.04 ro 18.04 went super good..with below command $sudo update $sudo upgrade $sudo do-release-upgrade After this you have to bunch of Y's to continue clean install (sorry for putting short cut process here). As i faced crazy issue after that.. I tried to install another toolkit in Ubuntu 18.04 and in process somehow it deleted some systemfiles as part of python3 removeal process... This step made me to spend almost 6-8hrs of non-stop hard hitting to fix 1. First my KeyBoard and Mouse stopped working after restart 2. Had to reboot and during BIOS coming up - I had to hit "Esc" key to go to GRUB GUI 3. I selected "recovery-mode" Linux image 4. Then selected "root" shell and try to install "sudo apt-get install ubuntu-desktop" and it failed crazy because Wifi not connected at this point and "$sudo update" failed 5. then I have to use "nmcli" commands to connect to my wifi (google nmcli commands) 6. After sucessfully connecting to wifi - i was able to run "sudo apr-get update && sudo apt-get upgrade" and still "sudo apt-get install ubuntu-desktop" was failing 7. But, keyboard started working (Hope I am not missing something b/n Step 6 & 7 - w.r.t isntalling gonme package with apt-get) 8. BIGGEST challenge was "Fail to start session" when trying to login to system (THIS WAS THE MOST TIME TAKEN HARD PROBLEM) 9. I had to so much of circus to fix this - this time once I got to login screen - I entered command promopt/terminal with "CTRL + ALT + F2" 10. First tried to install "sudo apt-get install ubuntu-package" - it said I dont know what are you talking about "ubunut-package" not found 11. Then I tried to install python3 (which is the main culprit for all this hassle) - started throwing dpgk: error not able to proceed (subprocess dpkg failed etc. something like that) 12. Even "sudo dpkg --configure -a" was not working 13. Then I installed "sudo apt-get intall aptitude" 14. sudo aptitude install python3-pip 15. sudo aptitude install python3-dev (not 100% sure of this - nothing harm to in reinstalling) 16. and etc. packages - all the once which are through error by "sudo apt-get install ubuntu-desktop" in previous steps 17. During 14, 15 and 16 steps - it will ask you to give your decision on what each step/command is going to do (if you feel all looks good- you can hit 'y' or Enter) Now you should not see any dpkg errors and reboot your system and you can happily login and recover all your hard work put in this silly life so far.. Have fun....!!!! with lot of bugs and blockers - which have simple fixs.... Kudos to Communities trying to help each other....

OpenVINO DL Streamer with GStreamer - Azure Marketplace app

noreply@blogger.com (psakamoori) — Mon, 20 Jul 2020 05:56:00 +0000

Deep Learning (DL) Streamer is a component of OpenVINO™ Toolkit that offers a streaming analytics framework based on GStreamer* multimedia framework for creating complex media analytics pipelines using OpenVINO Inference Engine. It provides optimal pipeline interoperability and optimized inferencing across Intel® architecture, CPU, iGPU and Intel® Movidius™ VPU.

This container provides Intel® distribution of OpenVINO™ Toolkit with DL Streamer and OvdlStreamer app that provides a configurable pipeline for object detection, tracking and classification. The app can take multiple input streams, perform object detection, tracking and classification using pre-trained models required by the OpenVINO™ toolkit Inference Engine, and send the results to Azure IoT Cloud and/as RTSP out.

Click Here to Download OpenVINO DL Streamer - Azure IoT Edge Module

Image Semantic Segmentation with OpenVINO 2020.R2

noreply@blogger.com (psakamoori) — Mon, 20 Jul 2020 05:45:00 +0000

The One Hundred Layers Tiramisu:Fully Convolutional DenseNets for Semantic Segmentation

Github (OpenVINO app code) to accelerate inference of FC-DenseNet-103 on Intel HW. Best performance can be seen on Cascade Lake Xeon CPU

OpenVINO Application code
OpenVINO FC-DenseNet-103 IR model files (FP32/FP16): IR Files

Note: IR files are generated from pre-trained TF FC-DenseNet-103 model file (with accuracy ~.60)

If you want to re-train: You can use Keras train.py

Age and Gender Classification using OpenVINO 2020.R1

noreply@blogger.com (psakamoori) — Sun, 22 Mar 2020 01:57:00 +0000

OpenVINO Ready To Deploy AI Vision Module

noreply@blogger.com (psakamoori) — Sun, 22 Mar 2020 01:56:00 +0000

OpenVINO - Train and Deploy Neural Network (AI Model) in seconds onto IoT Edge device

noreply@blogger.com (psakamoori) — Thu, 20 Feb 2020 06:34:00 +0000

Lets look at the challenges being faces by AI developers in training and deploying AI model and how OpenVINO and Azure IoT helps solve the problem

Challenges in training and deploying AI model

Choosing a Neural network model
Train/re-train till model converge (Costly and time consuming task)
Deploying it on Edge device (IoT device's / Laptop / Desktop )

Prerequisites/System requirements - Free Subscription, Software, Hardware and setup:

Prerequisites

Subscription:

Hardware (Any two of the below):

Intel Powered IoT Device (Ex: UP2 AEEON IOT Edge device)
Intel Powered Laptop/Desktop with Core i7/i9
Neural Compute Stick (USB powered) acts as AI accelerator at Edge
USB Web cam (Ex: Logitech )

Operating System and Software (pre-installed on Host Machine):

Ubuntu 16.04
Intel® Distribution of OpenVINO™ toolkit
Azure IoT Edge Runtime
Docker

Setup:

Host machine should be connected to Azure IoT Hub
Enable xhost (ubuntu terminal command: $sudo xhost +)

Solution:

This Azure Marketplace (deploy) will make total end2end train and deploy onnx model in less than a minute. Application uses docker image built on OpenVINO with ONNX Run-time execution provider (EP)

Details: How it works:

Step 1: Training using customvision.ai in three simple steps

Login to customvision.ai - Upload few train samples (min 25)
Annotate them
Do a quick train

Ref: Getting started with customvision

Step 2: Deploying OpenVINO AI Vision Module on to IoT Edge device

Click "Get It Now" on Azure Marketplace
Select device from IoT Hub to deploy
Once deploy is successful - will see "OpenVINOReadyToDeployAIVisionModule" Edge module running
Expected output: Camera stream rendering on to display

Step 3: Passing ONNX model to the app with "Twin Updates"

1.Copy "ONNX model URL from" -> customvision.ai -> select project -> Performance -> Export -> Click on ONNX -> Copy ONNX model URL
Goto "portal.azure.com" -> IoT Hub -> IoT Edge -> Select device
Click on "setmodules" -> click on -> "OpenVINOReadyToDeployAIVisionModule"
Selecting "Twin Module Settings" -> pass "ONNX model URL path selected in step 3.1" to -> inference_files_zip_url (looks like inference_files_zip_url="onnx url path")
Finally click "Update" and "Review+Create"
Expected output: OpenVINO app will restart the stream and starts running inference based on the ONNX model passed (Object detection/Image classification) (Camera: Should be pointing to object/image of interest to do recognition/classifciation) Note: If no NCS2 connected, inference will start on Intel CPU

DEMO VIDEO FOR YOUR REFERENCE

Setup is one time process - need some patients to go through cloud setup (if doing it first time) - Happy to answer any question - leave a comment
Deploying (Docker pull) application will take decent time (only once per device) based on network speed

Note: Working on making light weight docker

Anaconda Installation on Mac with Python 3.7

noreply@blogger.com (psakamoori) — Mon, 07 Oct 2019 01:49:00 +0000

Q&A on Deep Learning concepts

noreply@blogger.com (psakamoori) — Wed, 17 May 2017 17:03:00 +0000

1. What is Deep Learning?

Deep learning is a subset of machine learning that is concerned with neural networks.

Deep learning represents a learning algorithm that learns representations of data through the use of neural nets.

2. What is Mean?

Mean: Average of all the numbers

3. What is Variance and Standard Deviation?

Variance: The variance (σ2) is a measure of how far each value in the data set is from the mean. The Average of Squared difference from the mean

Standard Deviation: Gives how spread out the numbers are

4. What is Perceptron?

A perceptron (type of neuron) takes serval binary inputs x1, x2,….x1,x2..and produces a single binary output

5. What is Sigmoid?

Just like perceptron, sigmoid (type of neuron) has inputs x1,x2….But instead of just being 0 or 1 these inputs can also take values between 0 and 1. Also, just like a perceptron, the sigmoid neuron has weights for each input w1, w2…and an overall bias. But out is not 0 or 1, instead σ(w.x + b), sometimes call logistic function.

6. What is Gradient?

Gradient is another word for "slope". The higher the gradient of a graph at a point, the steeper the line is at that point. A negative gradient means that the line slopes downwards.

7. Why “Gradient Descent”?

We learned that sigmoid takes Input X (ranging from 0 to 1), W weights and B bias values to compute the output of a neuron. But, in order to calculate “W” and “B” we need a function to calculate them. So, Gradient Descent is one of the methods to calculate W and B values.

8. Explain “Gradient Descent” and “Stochastic Gradient Descent (SGD)”?

Both algorithms are methods for finding a set of parameters that minimize a cost/loss function by evaluating parameters against data and then making adjustments.

In “Gradient Descent”: You will evaluate all the training samples for each set of parameters

In “SGD”: You will evaluate 1 training sample for the set of parameters before updating them.

Helps find which “Weights” and “Bias” number results in minimizing cost function

C(w,b) = 1/2n∑ ||y(x) – a||²

w-> Weight, b->bias

y(x) -> What the output (ref output) should be for input x

a -> Output given by network for a given ‘x’, ‘w’ and ‘b’

9. Explain “Training” a CNN model?

Supervised -> labeling the data that model needs to be trained on

Unsupervised Learning -> Training without labeling

10. What is an epoch?

Epoch -> is a single pass through entire data set.

11. Explain “Back propagation”?

An expression for the partial derivative of the Cost function (C) w.r.t to any weight (or bias).

The expression tells us how quickly the cost changes when we change the weights and biases.

The goal of backpropagation is to compute the partial derivatives ∂C/∂w and ∂C/∂b of the costfunction C with respect to any weight w or bias bb in the network

Four Fundamental Equations behind BP

Backpropagation is about understanding how changing the weights and biases in a network changes the cost function. Ultimately, this means computing the partial derivatives ∂C/∂wljk, and ∂C/∂blj.

We first introduce an intermediate quantity, δjl, which we call the error in the jth neuron in the lth layer.

Backpropagation will give us a procedure to compute the error δjl, and then will relate δlj to ∂C/∂wjkl and ∂C/∂bjl.

“BP” is about Understanding how changing weights and biases in a network changes the cost function (ex: Output of Gradient Descent/SGD). Ultimately this mean computing the partial derivatives of cost function w.r.t

The method calculates the gradient of a loss function with respect to all the weights in the network

12. Examples “Cost Functions”?

“Gradient Descent”, “Stochastic Gradient Descent (SGD)”, L1 Regularization, and L2 Regularization

13. What are “Activation Functions”?

Softmax: Its similar to Sigmoid ,

but with different function

14. What are hyper parameters and How to select “Hyper Parameters”?

Learning rate, epoch(s), batch size (how many samples to process per iteration of training)

Regularization Parameter

15. How to initialize “weights” and “biases” for given network/model? why?

Generally “Weights” and “Biases” are initialized randomly using Gaussian Distributions with Mean=0 and Standard Deviation=1

Why?? Ex: if you have 1000 input nodes and assume that you zeroed 500 samples and for reaming samples standard deviation is √501 (500 samples + 1 bias) => 22.4…as shown below that is z (w.x+b) has a very bad Gaussian Distribution, not sharply peaked

16. When do you stop “training”?

Neg case: If cost not going down (or) test accuracy is not improving (or) during overfitting (Cost decrease but Accuracy doesn’t improve)

Positive Case: If desired accuracy is reached

17. What is Overfitting? How to avoid it?

When Network shows decrease in cost, but accuracy does not change as expected w.r.t decrease in cost. In this scenario we say network is overfitting or overtraining.

We need to check how Cost of train and test data varying as the network learns. If you see opposite scenario (Cost of Training set going down and Cost of Test set going up) it’s a indication of Overfitting (Network not learning from dataset)

Another sign is classification of training data set. If accuracy on training data set is 100% and test data set is ~80%. It means that network memorizing training samples (not actually learning it features).

The obvious way to detect overfitting is to use the approach above, keeping track of accuracy on the test data as our network trains. If we see that the accuracy on the test data is no longer improving, then we should stop training.* It might be that accuracy on the test data and the training data both stop improving at the same time. Still, adopting this strategy will prevent overfitting.

Increasing the amount of training data is one way of reducing overfitting – Not always possible
Using Regularization techniques, weight decay or L2 regularization. The idea of L2 regularization is to add an extra term to the cost function, a term called the regularization term

18. What is “dropout”?

“Dropout” is another technique to remove “Overfitting”. It does not change Cost function as done by L1 and L2, instead it changes the network itself.

During training we forward-propagate input ‘x’ through the network and then back propagate to compute the “gradient”. With dropout, this process is modified. We start by randomly removing hidden neurons in the network, while leaving the input and output neurons untouched and we repeat this process by randomly adding back previously removed hidden neurons and removing other mini-batch of hidden neurons. We keep repeating this process and calculating gradient and updating weights & biases in the network.

19. What is L1 and L2 norm?

L1-norm is also known as least absolute deviations (LAD), least absolute errors (LAE). It is basically minimizing the sum of the absolute differences (S) between the target value (Yi) and the estimated values (f(xi)):

L2-norm is also known as least squares. It is basically minimizing the sum of the square of the differences (S) between the target value (Yi) and the estimated values (f(xi):

20. Difference between L1 and L2 Regularization?

L1 regularization shrinks the weight much less than L2 regularization does

21. Explain “Precision” and “Recall”? (Ex: Face Detection)

TP -> Correctly detected faces

FP -> detecting non-face as faces

FN -> No.of.faces missed

Precision -> Percentage of identified faces that are correct (TP)/ ((TP + FP))

Recall -> Percentage of correctly detected faces to the total no of faces (TP)/ (TP + FN))

22. Formula to compute size of Layers?

Output size = [((Width – Kernel Size + 2*Padding)/Stride ) + 1]

23. How to calculate no.of.operations per layer (FLOPS)?

(Kernel W X Kernel H X Output No.of features X No.of channels of input image X Image W X H)

24. How would you handle an imbalance data set?

Try collecting more data to even the imbalances

Resample the dataset to correct the imbalance

Try different algorithm altogether on your dataset

25. When should you use “classification” over “regression”?

Classification produces discrete values and dataset to strict categories, while regression gives you continuous results that allow you to better distinguish differences between individual points. You would use classification over regression if you wanted your results to reflect the belongingness of data points in your dataset to certain explicit categories (ex: If you wanted to know whether a name was male or female rather than just how correlated they were with male and female names.)

Please leave a comment, if have any questions on above Q&A (or) Want to ask any new Question on this topic of Deep Learning.

Happy Reading.

How to parse caffemodel file

noreply@blogger.com (psakamoori) — Sat, 29 Apr 2017 05:26:00 +0000

Written a python script which helps to parse "weights" and "biases" of each layer from a pretrained ".caffemodel" file

github : CaffeModelFileParser

Leave a comment, if any questions

Caffe2 - Deep Learning framework

noreply@blogger.com (psakamoori) — Sat, 29 Apr 2017 05:21:00 +0000

Caffe is Machine Learning framework from BVLC

Caffe2 is newer release from BVLC, A New Lightweight, Modular, and Scalable Deep Learning Framework

Website : caffe2 , github

Leave comment, if any questions

Python code to convert jpg to rgb format

noreply@blogger.com (psakamoori) — Thu, 01 Sep 2016 00:29:00 +0000

Below utility will help convert
jpeg file to rgb file format

Any question leave a comment.

Save the code to a file "jpg2rgb.py"

How to use: python jpg2rgb.py

import cv2
import os, sys
import numpy as np
import argparse

def convert_jpg2rgb(ipfile, opfile):

# Decoding input jpeg file
ipimg = cv2.imread(ipfile, cv2.IMREAD_COLOR)

# Shape of input image
w, h, d = ipimg.shape

# Reading clip name
s = ipfile.split('\\')
clip_name = str(s[len(s)-1])

s = clip_name.split('.')
clipName = str(s[0])

# Reading B, G, R buffers
b, g, r = cv2.split(ipimg)

opimg = open(opfile + '/' + str('rgb_') + str(clipName) + str('.rgb'), 'wb')

for i in range(0, h):
for j in range(0, w):
opimg.write(r[i,j])
opimg.write(g[i,j])
opimg.write(b[i,j])
opimg.close()

return True

def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('ip_jpg', help="Path to Input jpg file")
parser.add_argument('op_rgb', help="Path to save rgb file")

args = parser.parse_args()
return args

if __name__ == "__main__":

args = parse_args()

ip_jpg_file = args.ip_jpg

if not os.path.isfile(ip_jpg_file):
print "jpg file Not found\nExiting..."
sys.exit()

op_rgb_file = args.op_rgb
if not os.path.isdir(op_rgb_file):
print "Invalid path\nExiting..."
sys.exit()

if(convert_jpg2rgb(ip_jpg_file, op_rgb_file)):
print "Conversion Sucessful"
else:
print "Conversion Failed"

Face Recognition - Machine Learning - Deep Neural Networks

noreply@blogger.com (psakamoori) — Fri, 26 Aug 2016 00:24:00 +0000

Face Recognition is very hard problem to solve using traditional CV techniques considering hard real world scenarios lighting, camera angle etc

Far better accuracy can be achieved in Face Recognition using Deep Neural Networks called Openface an opensource free software from Carnegie Mellon university

I have been working on this for couple of months now and very impressed with the outcome

To Try it out -

Follow these steps

Introduction - https://cmusatyalab.github.io/openface/
Code - https://github.com/cmusatyalab/openface
Torch - Frameworks - http://torch.ch/docs/getting-started.html#_
Training and classification - https://cmusatyalab.github.io/openface/demo-3-classifier/

System requirements
---------------------------

Linux 12+
Python 2.7+ with additional dlib, sckit packages
CudNN drivers - If using nvidia GPU

Any questions with setup or how to ..
please leave a comment will help you.

Computer Vision with MATLAB for Object Detection and Tracking

noreply@blogger.com (psakamoori) — Tue, 30 Apr 2013 05:35:00 +0000

View this email as a web page.
To ensure our emails reach your inbox, add the domain @em.mathworks.com to your safe sender list. Learn more.

Dear Pradeep Sakhamoori

MathWorks India invites you to a complimentary webinar:

Computer Vision with MATLAB for Object Detection and Tracking

8 May 2013
3:00 PM IST (India Standard Time)

Computer vision uses images and video to detect, classify, and track objects or events in order to understand a real-world scene. In this webinar, we dive deeper into the topic of object detection and tracking. Through product demonstrations, you will see how to:

Recognize objects using SURF features
Detect faces and upright people with algorithms such as Viola-Jones
Track single objects with the Kanade-Lucas-Tomasi (KLT) point tracking algorithm
Perform Kalman Filtering to predict the location of a moving object
Implement a motion-based multiple object tracking system

This webinar assumes some experience with MATLAB and Image Processing Toolbox. We will focus on the Computer Vision System Toolbox.

About the Presenter: Bruce Tannenbaum works on image processing and computer vision applications in technical marketing at MathWorks. Earlier in his career, he developed computer vision and wavelet-based image compression algorithms at Sarnoff Corporation (SRI). He holds an MSEE degree from University of Michigan and a BSEE degree from Penn State.

Invite a Colleague

MATLAB Digest - March 2013

noreply@blogger.com (psakamoori) — Tue, 19 Mar 2013 17:06:00 +0000

Overcome your signal line frustration with smart signal routing in Simulink. View this issue online.

March 2013

Call for Papers: MATLAB EXPO 2013 India

noreply@blogger.com (psakamoori) — Fri, 01 Mar 2013 06:07:00 +0000

You are invited to submit papers for MATLAB EXPO 2013 India.

Wednesday, 21 August 2013
NIMHANS Convention Centre, Bangalore

This annual one-day event brings together engineers and scientists from leading organizations to learn about the latest product capabilities in MATLAB and Simulink, exchange ideas, share successful use cases, and network with MathWorks experts, partners, and industry peers.

All accepted papers will appear on the event proceedings web site and in the MATLAB EXPO handbook published by MathWorks, giving you an excellent opportunity to share your experience with using MATLAB and Simulink to solve your project challenges.

The deadline to submit abstracts is 15 April. For more submission details, deadlines, and suggested areas of focus, please visit matlabexpo.in.

We look forward to seeing you at MATLAB EXPO 2013 India.

Sincerely,
The MathWorks India Event Team


	Important Dates

	•		15 April: Deadline to submit abstracts
	•		06 May: Selection and intimation to speakers
	•		1 August: Final papers and presentations due
	•		21 August: Date of MATLAB EXPO

Visit matlabexpo.in for the latest EXPO updates and to see video proceedings from the 2012 event.

Follow the Event

#MATLABEXPO

Control System Design, Analysis and Real-Time Testing

noreply@blogger.com (psakamoori) — Thu, 29 Nov 2012 19:18:00 +0000

Control System Design, Analysis and Real-Time Testing using MATLAB and Simulink

11 Dec 2012
3:00 PM IST (India Standard Time)

In this webinar, you’ll learn how MATLAB & Simulink® are utilized in the development of an embedded control system including implementation and testing on hardware. Our demonstration will emphasize how to design, simulate and test a complex system that incorporates multiple domains such as mechanical, electrical and hydraulic – that are typically isolated across different software platforms and thus not simulated in a common framework.

Starting from underlying physical principles, we will build models that can expand and accelerate the development process through the use of simulation. You will see how you can simulate the controller and the plant together, optimize the control system, and generate code for HIL testing—all before building a prototype. We will demonstrate the integration of simulated and real components to enable Hardware-in-the-Loop (HIL) testing in support of early Verification and Validation (V&V) against top-level requirements.

You’ll learn how to:

Build models of mechanical, electrical and control software component
Employ models to optimize component designs against system requirements
Employ models to support the production and testing of integrated systems

About the Presenters:

Terry Denery is an expert in Modeling. He runs hundreds, if not thousands, of simulations to pursue the best design of electrical, mechanical, and control systems.

Sam Mirsky’s forte is working with the hardware; these two skill sets complement each other perfectly. Using MathWorks tools Sam can rapidly prototype Terry’s best design, and prove whether it will really work or not. This could not be done without good modeling and rapid conversion of these models into real hardware systems. Together, Terry and Sam will show how to do this in MATLAB & Simulink, the platform for Model-Based Design.

Q&A Session: Prasanna Deshpande – Application Engineer, MathWorks India

Invite a Colleague

Using HDL Coder and HDL Verifier for FPGA and ASIC Designs

noreply@blogger.com (psakamoori) — Fri, 16 Nov 2012 05:37:00 +0000

Dear Pradeep Sakhamoori,

MathWorks India invites you to a complimentary webinar:

Using HDL Coder and HDL Verifier for FPGA and ASIC Designs

27 Nov 2012
3:00 PM IST (India Standard Time)

In this webinar you will learn how you can leverage our HDL Code Generation and Verification products to accelerate your FPGA design cycle and avoid costly mistakes. Using HDL Coder you can prototype your algorithm on FPGAs or implement it on ASICs and FPGAs directly from Simulink. With HDL Verifier, you can co-simulate your HDL code with ModelSim and perform FPGA based accelerated simulations.

MathWorks engineers will demonstrate the latest enhancements to HDL Coder, which generates synthesizable Verilog® and VHDL® code from Simulink models, MATLAB code, and Stateflow charts.

We will discuss the following topics:

Simulink system level design
Verilog and VHDL code generation using HDL Coder
Optimization techniques for efficient FPGA implementation
Pipelining and resource sharing
Co-simulation with ModelSim
Programming your HDL code on FPGA boards
FPGA-in-the-Loop verification

Presenter: Stephan van Beek – Application Engineer, MathWorks

Q&A Session: Puneet Kumar – Principal Application Engineer, MathWorks India

Invite a Colleague

Production Code Generation with Simulink and Embedded Coder

noreply@blogger.com (psakamoori) — Thu, 18 Oct 2012 03:33:00 +0000

View this email as a web page.
To ensure our emails reach your inbox, add the domain @em.mathworks.com to your safe sender list. Learn more.

Production Code Generation with Simulink and Embedded Coder

30 Oct 2012
3:00 PM IST (India Standard Time)

In this webinar, you’ll learn how to generate high-quality and optimized C/C++ code from Simulink models for deployment in production systems. You’ll also learn how to build, profile, and verify generated code on target processors.

Featuring MATLAB and Simulink’s New Look and Feel in Release R2012b

Topics include important new technologies in latest releases including:

Target optimized code for software deployment
Function prototype control to ease software integration
Float- to Fixed Point conversion advancements
Bidirectional model and code traceability
Software- and Processor-in-the-Loop (SIL and PIL) testing
Support for AUTOSAR, DO-178, IEC 61508, ISO 26262, and MISRA-C

This webinar is primarily targeted at people new to Model-Based Design and production code generation, but will also be of interest to those already applying Model-Based Design.

A Q&A session will follow the presentation and demos.

About the Presenter: Tom Erkkinen is product manager for code generation and certification products at MathWorks. He has 20 years industry experience using Model-Based Design in aerospace and automotive industries

Q&A Session: Shobhit Shanker, Application Engineer, MathWorks India

Invite a Colleague

We look forward to seeing you online on Tuesday, 30th of October, 2012.

Regards,

Webinar Team
MathWorks India

Openings for Android, DTV and STB professionals - Hyderabad

noreply@blogger.com (psakamoori) — Thu, 04 Oct 2012 11:11:00 +0000

Openings for Android, DTV and STB professionals

Area	Technology	Req
Android	Stability	4
	Triage	4
	Build and Release	2
Android MM FW	Audio	3
	Video	2
	Display	2
	Graphics	1
Driver	Audio	1
	Video	1
	Display	1
	Graphics	1
DTV MW Integration	DVB,ATSC,ISDB,AV-Sync	3
DTV FW	Transport Stream, Video Frameworks, SBT, G Streamer	3
STB Drivers	Audio	1
	Video	1
	Display	1
	HDMI	1

Send your resume to pradeep.sakhamoori@smartplayin.com

High-Performance Image Processing in MATLAB

noreply@blogger.com (psakamoori) — Thu, 27 Sep 2012 13:55:00 +0000

The new MATLAB Desktop and Simulink Editor are available in MATLAB and Simulink R2012b. View this issue online.

September 2012

Designing and Implementing Signal Processing, Video Processing and Communication Systems using MATLAB and Simulink

noreply@blogger.com (psakamoori) — Tue, 28 Aug 2012 13:17:00 +0000

Designing and Implementing Signal Processing, Video Processing and Communication Systems using MATLAB and Simulink

11 Sep 2012
10 am to 2.45 pm
Sapphire Ballroom, Hyderabad Marriott Hotel & Convention Centre, Hyderabad

Designing complex, next generation signal processing, video processing and communications products in shortened design cycles magnifies the need for continuity in design methods. Gaps in design flows result in sub-optimal system designs, negatively impact project schedules and ultimately deplete project budgets.

Join us for this free seminar to learn about how the new capabilities in MATLAB and Simulink – like System Toolboxes in MATLAB, automatic C code generation from MATLAB, and HDL code generation and FPGA-in-loop verification using Simulink – can help engineers improve overall design flow efficiency and signal processing and communications systems performance.

MathWorks engineers will use several examples to demonstrate the following:

Simplify the system design process by using high-level function libraries provided by System Toolboxes in MATLAB
Run efficient system simulations by leveraging stream processing to reduce scheduling overhead and speed up simulations
Reuse algorithmic level system designs created in MATLAB and Simulink to analyze the effects of hardware constraints on data types by converting floating-point algorithms to fixed-point and running verification suites
Reduce design cycle time through early prototyping and maintain design flow continuity by using automatic C and HDL code generation technologies
Reduce verification effort by reusing test benches from the algorithmic stage through to hardware-in-loop / FPGA-in-loop stage

Who Should Attend

Target audience:This seminar is recommended to anyone designing or implementing Signal Processing, Video Processing and Communications Systems on DSPs and FPGAs

Algorithm Designers
System Engineers
Signal Processing Engineers
Video System Engineers
Communications System Engineers
Embedded Software Engineers
FPGA Designers
Test and Verification Engineers

Prior knowledge of MathWorks products is not required.

Please note this Seminar is not designed for students. If you are a student, please do not register for this seminar.

Invite a Colleague

Agenda
09:30- 10:00	Registration
10:00-10:15	Welcome Address
10:15-11:30	Addressing System Design Challenges with Model-Based Design Signal Processing, Video Processing and Communication System Design with MATLAB and Simulink MATLAB Capabilities for System Design Modeling and Simulation Handling Large Datasets
11:30 -12:00	Break
12:00 -13:30	From MATLAB to C : Bridging the Gap Between Algorithm Design and Implementation Fixed Point Design in MATLAB Automatic C-code Generation from MATLAB Optimizing the Generated Code for Target Platforms Implementing Algorithms on FPGAs and ASICs Automatic HDL Code Generation from MATLAB, Simulink, and Stateflow Integrated HDL Verification FPGA-in-the-Loop Verification
13:30 -13:45	Wrap up
13:45 – 14:45	Lunch

Invite a Colleague

MATLAB to C Made Easy

noreply@blogger.com (psakamoori) — Fri, 08 Jun 2012 04:08:00 +0000

MATLAB to C Made Easy

21 Jun 2012
3:00 PM IST (India Standard Time)

In this webinar we demonstrate the workflow for generating readable and portable C code from your MATLAB algorithms using MATLAB Coder. Using the command line approach or the graphical project management tool, you can introduce implementation requirements to you algorithms written in MATLAB and generate readable source code, or a standalone compiled executable or a library that can be shared across your organization.

MathWorks engineers also explore how you can automatically generate MEX functions that can be used to verify the behavior of the generated code back in MATLAB or to accelerate computationally intensive portions of your MATLAB code by running it at compiled speed.

Invite a Colleague

We look forward to seeing you online on Thursday, the 21st of June, 2012.

Regards,

Marketing Programs
MathWorks India