Learn About Me
For most uptodate information about me, please visit my LinkedIn Page
I am a Ph.D. candidate in computer science under the supervision of Prof. Bhaskar Krishnamachari, and I expect to graduate in May 2024. My research interests include machine learning, data-driven algorithms, anomaly detection, and edge computing. As a research assistant at the ANRG lab of USC, I work on developing and evaluating machine learning models for detecting and analyzing various attacks on IoT networks, such as DDoS and Sybil attacks. I have over four years of experience in this field, and I have published several papers in top-tier conferences and journals.
Previously, I was a software engineer intern at Google, where I worked on designing NLP/ML based systems to automatically generate ads and campaigns. I developed and integrated a prompt-engineered 24B LLM in production for generating search ads keywords tailored to advertisers' queries, resulting in high-quality and relevant keywords. I also designed and evaluated a prompt-engineered and fine-tuned 24B LLM on the ads help articles for addressing advertisers' general questions, achieving high relevance in addressing advertiser inquiries.
I am passionate about applying machine learning to solve real-world problems. I am always eager to learn new technologies and techniques, and to collaborate with other engineers and researchers. I am looking for opportunities to further advance my skills and knowledge, and to contribute to cutting-edge projects in these fields.
Education
I have used innovative AI based methods for detecting anomalies and also designed data driven algorithms for making informed decisions.
Aug. 2019 - Present
Developed an optimal computation offloading algorithm
for a general Markovian communication channel. I have also worked on efficient offloading by using neural networks,
Hidden Markov Models, reinforcement learning and other machine learning techniques.
Sep. 2017- Jun. 2019
Developed multiple OpenDaylight controllers in software defined networks in order to
control the TCP/UDP generated traffics in the network created by Mininet.
Sep. 2013 - Jul. 2017
Sep. 2009 - Jun. 2013
Work Experience
May 2023 - Aug. 2023
Team: Ads Conversational AI - Designing NLP/ML based systems to automatically generate Ads and campaigns:
• Developed and integrated a prompt-engineered 24B LLM in production for generating search ads keywords tailored to advertisers' queries, resulting in ∼90% relevance and high-quality keywords suitable for production use.
• Designed and evaluated a prompt-engineered and fine-tuned 24B LLM on the ads help articles for addressing advertisers' general questions; adopted the fine-tuned model, achieving ∼90% relevance in addressing advertiser inquiries.
• Developed and integrated a prompt-engineered 24B LLM in production for generating search ads keywords tailored to advertisers' queries, resulting in ∼90% relevance and high-quality keywords suitable for production use.
• Designed and evaluated a prompt-engineered and fine-tuned 24B LLM on the ads help articles for addressing advertisers' general questions; adopted the fine-tuned model, achieving ∼90% relevance in addressing advertiser inquiries.
May 2022 - Aug. 2022
Team: Responsive Search Ads - Improving the quality of the automatically generated text ads.
• Implemented workflows/dashboard for continuous evaluation of non-EN text ads quality.
• Improved quality of automatically generated text ads through heuristic and ML modeling solutions:
- Optimized title-based headline extractor filters, increasing the selection of high-quality headlines by ∼16%.
- Improved mismatched language filter model accuracy by ∼93%, significantly reducing text ads rejections.
- Implemented an ML workflow to aggregate and analyze data for enhancing the grammaticality model of non-EN text ads.
• Implemented workflows/dashboard for continuous evaluation of non-EN text ads quality.
• Improved quality of automatically generated text ads through heuristic and ML modeling solutions:
- Optimized title-based headline extractor filters, increasing the selection of high-quality headlines by ∼16%.
- Improved mismatched language filter model accuracy by ∼93%, significantly reducing text ads rejections.
- Implemented an ML workflow to aggregate and analyze data for enhancing the grammaticality model of non-EN text ads.
Jun. 2016 - Sep. 2016
I worked with Iotivity standard, one of the most important standards defined for the Internet of Things. At the first step, I used Iotivity to connect to different sensors for sending commands and also receiving data from them totally remotely over the internet. The other focus of my research was on sensor information processing in order to learn the users’ behavior.
Research Experience
2019 - Present
• Distributed Denial of Service (DDoS) Attack Analysis:
- Synthesized and enhanced a real IoT time-series dataset with 350M samples, using real IoT network traffic distribution.
- Developed tunable DDoS attacks on IoT nodes with different start times, duration, and traffic patterns.
- Designed DDoS detection models using MLP, CNN, LSTM, AEN, and GCN with F1-score up to 91%
- Designed and evaluated both prompt-engineered and fine-tuned GPT-3.5-turbo for DDoS detection reasoning.
• Sybil Attack Analysis:
- Preprocessed and feature engineered a mobile encounter dataset with more than 3M samples.
- Designed Sybil attackers mimicking the behavior of real nodes using WGAN model
- Developed a GCN-based detection model with 91% F1-score, outperforming traditional MLP models by 10%
• Living Off The Land Binary (LOLBin) Malware Analysis:
- Implemented FastText, BERT, RoBERTa, and SBERT embedding for Unix command representation.
- Designed hybrid LOLBin detection/reasoning mechanism using LSTM/GPT-3.5-turbo with 94% F1-score
• University Digital Twin (Gemini):
- Created a unique students’ course registration dataset with more than 200K samples.
- Modeled epidemic spread on a school campus and developed simulations to analyze the impact of different policies.
- Designed and developed a greedy course scheduling algorithm, improved building entrance delay up to 3×.
- Led a group of two students developing a real-time simulation web app dashboard
• Administrating the ANRG Lab Linux servers
Sep. 2017 - Jun. 2019
Worked on Mobile Cloud Computing and Applied Machine Learning. Specifically, I’m working on Markov Chain, Markov Decision Process, Dynamic Programming, Machine Learning, Hidden Markov Model, Classification, Reinforcement Learning.
Oct. 2016 - Jul. 2017
Worked on software-defined networks (SDN). I used the Mininet network emulator to create a virtual SDN network for OpenDaylight (ODL) to control. I developed a topology which had several ODL controllers and could control the TCP/UDP traffic generated by the nodes in the network. I also worked on the Virtual Machines and tried to make a connection between them with a distributed virtual multilayer switch called “OpenVSwitch.”
Publications
Certificates
Honors and Awards
2017-2019
2017
2013
Selected Course Projects
Developed K Nearest Neighbor, Continuous Nearest Neighbor, and Geo-Social Query Processing
Developed neural network in Matlab to recognize Persian alphabets
Implemented residual networks in Keras in order to detect the happy faces.
Developed YOLO model in Keras.
Developed word embedding and LSTM units in Keras for adding emojies to the sentences according to their meanings.
Developed 1-D convolutional layers, GRU layers, and dense layers in Keras.
Developed Neural Machine Translation (NMT) model by using attention model.
Design and implementation of an object oriented model of a social network like LinkedIn in C++
Design and Implementation of a Secure File System in C
Implemented the simulation in Matlab
Technical Skills
Matlab, C, C++, Python
TensorFlow, Keras
Linux, Windows
LATEX, Microsoft Office
Mininet, OpenDayLight, NS-2
Verilog, Modelsim, Quartus
Raspberry Pi, BeagleBone
AVR, ARM
Pspice, Hspice, Advanced Design System, Multisim, Labview, Proteus, Codevision
Professional Activities
Teaching Experience
Spring 2020
Instructor : Prof. Andrew Goodney
Fall 2019
Instructor : Prof. Andrew Goodney
Spring 2019
Instructor : Prof. Terry Todd
Fall 2018
Instructor : Prof. Terry Todd
Spring 2018
Instructor : Prof. Terry Todd
Fall 2017
Instructor : Prof. Terry Todd
Spring 2017
Instructor : Prof. Vahid Shah-Mansouri
Fall 2018
Instructor : Prof. Amir Masoud Rabiei
Fall 2016
Instructor : Prof. Ali Afzali-Kusha
Fall 2016
Instructor : Prof. Leila Yousefi
Spring 2016
Instructor : Prof. Mohammad Neshat
Spring 2016
Instructor : Prof. Hossein Imaneini
Spring 2016
Instructor : Prof. Mohammadreza Kolahdouz
Fall 2015
Instructor : Prof. Nayereh Majd
Fall 2014
Instructor : Prof. Manouchehr MoradiSabzevar
Language Skills
Reading: 28/30, Listening: 29/30, Speaking: 24/30, Writing: 29/30
Verbal : 151/170, Quant: 168/170, Writing : 4.5/6.0
Selected Courses
Neural Networks and Learning Machines, Machine Learning, Deep Learning
Advanced Analysis of Algorithms, Analysis of Algorithms, Engineering Optimization, Engineering Probability and Statistics, Engineering Mathematics, Calculus I, II
Geospatial Information Management
Advanced Programming, Introduction to Computing Systems & Programming
Cloud Communication, Wireless Communication Networks, Computer Networks, Digital Communication Laboratory