Ahmad Beirami

Research Scientist

EA Digital Platform - Data & AI

Electronic Arts, Redwood City, CA, USA


Ahmad Beirami is a research scientist at Electronic Arts's EADP – Data & AI, where he leads fundamental research and development on training agents for game AI and playtesting. His research interests broadly include machine learning, information theory, signal processing, and statistics. Prior to joining EA in 2018, he held postdoctoral fellow positions at Duke, MIT, and Harvard. He received the BS degree in 2007 from Sharif University of Technology, Iran, and the PhD degree in 2014 from the Georgia Institute of Technology, in electrical and computer engineering. He is the recipient of the 2015 Sigma Xi Best PhD Thesis Award from Georgia Tech.

Professional Experience

Research Scientist @ EA Digital Platform - Data & AI (Apr 2018-present)

    • I lead fundamental research and development on training human-like and believable agents for playtesting and game AI in modern complex games. I draw upon and develop a variety of control and machine learning methods (including the state-of-the-art deep reinforcement learning) to solve these problems in a hierarchical and cost-efficient manner (Presented at NeurIPS 2018 Workshops and AAAI 2019 Workshops).

Postdoctoral Fellow @ Harvard SEAS (Mentor: Vahid Tarokh, Sep 2016-Dec 2017)

    • I derived an estimator of the out-of-sample loss, which is computable solely using the available data and closely tracks cross validation for a broad range of learning objective functions including Gaussian mixture models, linear/logistic regression, LASSO, and ridge regression, and neural networks. (NIPS 2017)
    • I provided a computationally efficient preprocessing step for drawing randomized features that generalize as well as kernel methods and scale as well as linear methods. My experimental results confirmed that the method is effective on a variety of regression and classification tasks. (AAAI 2018)

Postdoctoral Associate @ MIT EECS (Mentor: Muriel Médard, May 2015-Mar 2018)

    • I derived fundamental limits on an important element of information theory, called guesswork, using a geometric characterization. Using this characterization, I generalized "weak typicality" to be applicable to atypical events, and proved several new and stronger results about information-theoretic concepts, such as error probability in decoding.

Postdoctoral Associate @ Duke ECE (Mentor: Robert Calderbank, May 2014- Aug 2016)

    • I derived inachievability bounds and learning rates for a variety of supervised learning problems by bounding the mutual information between the data sample and the learning function, assuming that the data of interest lives on a low dimensional manifold.


Ph.D. in ECE @ Georgia Tech (Advisor: Faramarz Fekri, 2014)

  • I derived information theoretic limits of redundancy elimination from network data using universal prediction and unsupervised learning with applications to network data compression.
  • I developed a joint memorization and clustering algorithm, called network compression that uses side information to approach the Shannon limit, with low complexity of implementation and processing delay.
  • I experimentally demonstrated that around 35% traffic reduction is achieved on Internet traffic traces of 30 wireless users with the proposed method.
  • I proved that these improvements are largely preserved if the fraction of nodes employing network compression exceeds a certain critical value leading to a fundamental phase transition.

M.Sc. in ECE @ Georgia Tech (2011)

B.Sc. in EE @ Sharif University of Technology (2007)

Honors & Awards

  • Distinction in Teaching Award, Harvard University (2017)
  • Exemplary Reviewer, IEEE Transactions on Communications (2016)
  • 2015 Sigma Xi Best Ph.D. Thesis Award, Georgia Tech (2015)
  • 2013-2014 Graduate Research Assistant Excellence Award, School of ECE, Georgia Tech (2014)
  • Outstanding Research Award, Center for Signal and Information Processing, Georgia Tech (2014)
  • Outstanding Service Award, Center for Signal and Information Processing, Georgia Tech (2014)
  • Best Student Paper Nomination, 51st IEEE International Midwest Symposium on Circuits and Systems (2008)
  • Bronze Medal, 20th Iranian National Mathematics Olympiad (2002)



  • Y. Zhao*, I. Borovikov*, A Beirami*, J. Rupert, C. Somers et al., “Winning isn’t everything: Training human-like agents for playtesting and game AI,” preprint, submitted to IEEE Transactions on Games. *Equal contribution. [arXiv][poster - presented in part at AAAI 2019 Workshop on RL in Games and NeurIPS 2018 Workshop on RL under partial observability]
  • A. Beirami and F. Fekri, "Universal compression with side information from a correlated source," preprint, submitted to IEEE Transactions on Communications. [arXiv]

Journal Papers

  • S. Salamatian, W. Huleihel, A. Beirami, A. Cohen, and M. Médard, "Why botnets work: distributed brute-force attacks need no synchronization," IEEE Transactions on Information Forensics and Security, to appear, 2019 (TIFS 2019). [arXiv]
  • A. Beirami, R. Calderbank, M. Christiansen, K. Duffy, and M. Médard, "A characterization of guesswork on swiftly tilting curves," IEEE Transactions on Information Theory, vol. 65, no. 5, pp. 2850-2871, May 2019 (TIT 2019). [arXiv]
  • H. Mahdavifar, A. Beirami, B. Touri, and J. S. Shamma, "Global games with noisy information sharing," IEEE Transactions on Signal and Information Processing over Networks, vol. 4, no. 3, pp. 497-509, September 2018 (TSIPN 2018). [arXiv]
  • A. Beirami, M. Sardari, and F. Fekri, "Packet-level network compression: realization and scaling of the network-wide benefits," IEEE/ACM Transactions on Networking, vol. 24, no. 3, pp. 1588-1604, June 2016 (TON 2016). [arXiv]
  • S. Callegari, M. Fabbri, and A. Beirami, "Very low cost chaos-based entropy source for the retrofit or design augmentation of networked devices," Analog Integrated Circuits and Signal Processing, vol. 87, no. 2, pp. 155–167, May 2016 (ALOG 2016). [ResearchGate]
  • A. Beirami, M. Sardari, and F. Fekri, "Wireless network compression via memory-enabled overhearing helpers," IEEE Transactions on Wireless Communications, vol. 15, no. 1, pp. 176-190, January 2016 (TWC 2016). [ResearchGate]
  • A. Beirami and H. Nejati, "A framework for investigating the performance of chaotic-map truly random number generators," IEEE Transactions on Circuits and Systems II, vol. 60, no. 7, pp. 446-450, July 2013 (TCAS II 2013). [arXiv]
  • A. Beirami, H. Nejati, and W. H. Ali, "Zigzag map: a variability-aware discrete-time chaotic-map truly random number generator," Electronic Letters, vol. 48, no. 24, pp. 1537-1538, November 2012 (Elec. Lett. 2012). [ResearchGate]
  • H. Nejati, A. Beirami, and W. H. Ali, "Discrete-time chaotic-map truly random number generators: design, implementation, and variability analysis of the zigzag map," Analog Integrated Circuits and Signal Processing, vol. 73, no. 1, pp. 363-374, October 2012 (ALOG 2012). [arXiv]
  • H. Nejati and A. Beirami, "Theoretical analysis of the characteristic impedance in metal-insulator-metal plasmonic transmission lines," Optics Letters, vol. 37, no. 6, pp. 1050-1052, March 2012 (Opt. Lett. 2012). [arXiv]

Selected Conference Papers

  • I. Borovikov and A. Beirami, "From demonstrations and knowledge engineering to a DNN agent in a modern open-world video game," in Proc. of AAAI 2019 Spring Symposium on Combining Machine Learning with Knowledge Engineering, March 2019 (AAAI-Make 2019). [ResearchGate][slides]
  • P. Farajiparvar, A. Beirami, and M. Nokleby, "Information bottleneck methods for distributed learning," in Proc. of 56th Annual Allerton Conference on Communication, Control, and Computing, October 2018 (Allerton 2018). [arXiv][slides]
  • S. Shahrampour, A. Beirami, and V. Tarokh, "On data-dependent random features for improved generalization in supervised learning," in Proc. of The Thirty-Second AAAI Conference on Artificial Intelligence, pp. 4026-4033, February 2018 (AAAI 2018). [arXiv][slides]
  • A. Beirami, M. Razaviyayn, S. Shahrampour, and V. Tarokh, "On optimal generalizability in parametric learning," in Proc. of 2017 Advances in Neural Information Processing Systems, pp. 3455-3465, December 2017 (NIPS 2017). [arXiv][poster]
  • A. Rezaee, A. Beirami, A. Makhdoumi, M. Médard, and K. Duffy, “Guesswork subject to a total entropy budget,” in Proc. of 55th Annual Allerton Conference on Communication, Control, and Computing, October 2017 (Allerton 2017). [arXiv][slides]
  • M. Nokleby, A. Beirami, and R. Calderbank, "Rate-distortion bounds on Bayes risk in supervised learning," in Proc. of 2016 IEEE International Symposium on Information Theory, pp. 2099-2103, July 2016 (ISIT 2016). [arXiv][slides]
  • J. Zhu, A. Beirami, and D. Baron, "Performance trade-offs in multi-processor approximate message passing," in Proc. of 2016 IEEE International Symposium on Information Theory, pp. 680-684, July 2016 (ISIT 2016) [arXiv][slides]
  • H. Mahdavifar and A. Beirami, "Diffusion channel with Poisson reception process: capacity results and applications," in Proc. of 2015 IEEE International Symposium on Information Theory, pp. 1956-1960, June 2015 (ISIT 2015). [ResearchGate][slides]
  • Y. Ma, D. Baron, and A. Beirami, “Mismatched estimation in large linear systems,” in Proc. of 2015 IEEE Intl. Symp. on Information Theory, pp. 760-764, June 2015 (ISIT 2015). [arXiv][slides]
  • A. Einolghozati, M. Sardari, A. Beirami, and F. Fekri "Capacity of discrete molecular diffusion channels," in Proc. of 2011 IEEE International Symposium on Information Theory, pp. 723-727, July 2011 (ISIT 2011). [arXiv]

See my Google Scholar page for a complete list of publications.

Selected Talks

  • "Toward cost-aware machine learning," ITA, Michigan (2019) [slides]
  • "Powering games with Data & AI," MIT, Harvard, CMU, Boston University, Georgia Tech, Duke, Columbia, Caltech, USC, Michigan (2018-2019). [slides]
  • "On generalizability and scalability in learning systems," Technicolor AI Lab, IBM AI, Borealis AI Lab (2017-2018). [slides]
  • "The geometry of guesswork," Yale, UT Austin, Texas A&M, Rice, Harvard, CU Boulder, Texas A&M, Georgia Tech (2015-2016). [slides]
  • "Packet-level redundancy elimination: an information theoretic approach," MIT, Samsung (2014). [slides]

Teaching Experience

  • Section instructor for Harvard ES 156 — Signals and Systems (Spring 2017) (Rating: 4.79/5.00)
  • Recitation instructor for MIT EECS 6.02 — Intro to EECS II: Digital Communication Systems (Fall 2015) (Rating: 6.00/7.00)
  • Principal instructor for Duke ECE 587/STA 563 — Information Theory (Spring 2015) (Rating: 4.86/5.00)