Catalogo dei Corsi di studio

Gabriele Tolomei, Ph.D.
email: tolomei@di.uniroma1.it home: http://www.di.uniroma1.it/~tolomei/
ORCID ID: orcid.org/0000-0001-7471-6659

EDUCATION
National Scientific Qualification (ASN)
Role: Professore di Seconda Fascia (Associate Professor)
Settore Concorsuale (Academic Field): 01/B1 - Informatica (Informatics)
From - To: 7 August 2018 - 7 August 2024
Role: Professore di Seconda Fascia (Associate Professor)
Settore Concorsuale (Academic Field): 09/H1 - Sistemi di Elaborazione delle Informazioni (Information
Processing Systems)
From - To: 26 July 2018 - 26 July 2024
Ph.D. in Computer Science 01/2008 - 11/2011
Università Ca' Foscari Venezia, Italy
Date: 17 November 2011
Thesis Title: Enhancing web search user experience: from document retrieval to task recommendation
Supervisors: Salvatore Orlando and Fabrizio Silvestri
Main Results: Developed an algorithm to discover the set of user tasks (i.e., group of search queries
having the same latent need) from historical data stored in search engine logs. This solution performed
16% better than traditional techniques in terms of F1 score, and about 5% better than the very best
state-of-the-art method known at that time.
Most valuable results published in ACM WSDM 2011 (best paper runner up) and ACM TOIS.
M.Sc. in Computer Science (summa cum laude) 10/2002 - 04/2005
Università di Pisa, Italy
Date: 21 April 2005
B.Sc. in Computer Science 10/1999 - 10/2002
Università di Pisa, Italy
Date: 18 October 2002

RESEARCH EXPERIENCE
Associate Professor 09/2019 -
Sapienza University of Rome, Italy

Assistant Professor 07/2017 - 08/2019
Università degli Studi di Padova, Italy
From - To: 18 July 2017 -
Goals: Research activities on topics at the intersection of machine learning and computer security. Es-
tablishment of a multidisciplinary team focused on adversarial machine learning in collaboration with
Università Ca' Foscari di Venezia, Italy.

Projects:
Interpretability of machine learning models: Formulated the problem of finding the "best" (i.e., less
costly) perturbations of input features so as to switch the predictions output by an existing tree-
based ensemble classifier. An algorithm to solve the problem has been proposed and its validity
has been assessed on a real-world use case (i.e., online advertising). Results have been published in ACM KDD 2017 conference; an extended manuscript has been submitted to IEEE TKDE journal, and it is about to be submitted for second-round review.
Robustness of machine learning models: Definition of the problem of training machine learning
models that are insensitive to (i.e., robust against) input perturbations crafted by a malicious
attacker, inspired by the notion of non-interference that is typical of the computer security do-
main. Proposal of a solution which is validated on public datasets. Results have been and will be
submitted for review to the ACM CIKM 2019 and IEEE ICDE 2020 conferences, respectively.
CSRF attacks detection using machine learning: CSRF attacks are one of the main web security
threats. Supervised learning techniques have been used to train a prediction model (i.e., a binary
classifier) on a dataset of labeled HTTP requests, collected with a browser extension developed
ad hoc. The classifier outperforms any (heuristic-based) baselines, scoring F1 = 0.72. Results and
dataset have been published to the IEEE EuroS&P 2019 conference.
IoT advertising: Online advertising is possibly the most profitable Internet-based business model
yet it is still "limited" to traditional devices (i.e., PCs and smartphones). A new idea of advertising
has been sketched so as to extend Internet advertising business to emerging pervasive and ubiqui-
tous interconnected smart devices, which are collectively known as the Internet of Things (IoT).
Such a novel vision along with the challenges to be addressed are described in a manuscript
which appears in the IEEE Communications Magazine.
Fraud-free, verifiable advertising costs: Ongoing collaboration with the Bosch Research and Tech-
nology Center of Pittsburgh, PA, USA. This project aims to introduce a new model of online
advertising, which allows advertisers who are often victims of frauds (e.g., ad click inflation) to
verify the amount of money they spend on their campaigns charged by ad networks and publishers.
Research Scientist 06/2014 - 07/2017
Yahoo Labs, London, UK
From - To: 2 June 2014 - 14 July 2017
Goals: Improve the engagement of users with Gemini, the integrated Yahoo online advertising plat-
form. Promote "high quality" advertisements using measures of post-click satisfaction, which go beyond
traditional Click-Through Rate (CTR). Analyse large-scale datasets from distributed computing envi-
ronments and mine interesting patterns via statistical/machine learning solutions. Design, implement,
and test innovative prototypes into production buckets as response to internal challenges, in collabora-
tion with Product and Engineering teams. Delivery results both internally and externally (e.g., research
paper submissions to top conferences like SIGIR, KDD, CIKM, RecSys, WSDM, WWW, etc.).

Projects:
Accidental ad click discovery and discounting: Design and implementation of a data-driven method-
ology to detect "accidental" clicks on CPC advertisements shown on Yahoo s properties, currently
used in production. Proposal of a technique for discounting those clicks so to balance between
inevitable drop in revenue and long-term satisfaction of advertisers. Proposed solution has been
published to the International Journal of Data Science and Analytics (JDSA) and patented with
the US Patent and Trademark Office.
Ad quality score: Design and implementation of a mechanism to monitor and report to the adver-
tisers the performance of their ad campaigns running on the Yahoo Gemini platform. Successfully
tested on a pool of selected advertisers and patented with the US Patent and Trademark Office.
Ad feature recommendations: Design and implementation of a system which is able to suggest
actionable changes to ad landing pages so as to improve their quality perceived by users. The
approach has been published to the ACM KDD 2017 conference and patented with the US Patent
and Trademark Office.
Postdoctoral Research Fellow 01/2012 - 06/2014
Università Ca' Foscari Venezia, Italy
From - To: 13 January 2012 - 1 June 2014
Goals: Novel application of machine learning and data mining techniques to large scale, heterogeneous
data sources with the aim of improving the effectiveness of web search engines.

Projects:
Classification of web authentication cookies: Automatic discovery of authentication cookies from
those stored in web browsers using a supervised learning technique. Design of a (semi-) automatic
method to build a ground truth of authentication cookies. Evaluation of four state-of-the-art
solutions proposed to detect authentication cookies. Development of a binary classifier, which
outperforms existing solutions by increasing the overall F1 score from 14% up to 23%.
Trending topics vs. web search: Analysed the impact of Twitter trending entities on user search
behaviour. Time-series regression revealed that signals from Twitter are useful to predict Google
Hot Trends and Wikipedia page requests or edits about 60% of times. Results published in CIKM
2013 workshop, ASE/IEEE SocialCom 2013.
Task-oriented web search and recommendation: Developed a graph-based model of task-based user
search behaviour. Implemented the prototype of a task recommender system, which suggested
tasks to web users instead of "traditional" queries, with about 50% precision. Results published in
OAIR 2013 conference and the ACM TOIS journal (ACM 2013 Computing Reviews Notable
Article).
Research Assistant 01/2008 - 01/2012
ISTI-CNR, Pisa, Italy
From - To: 16 January 2008 - 8 January 2012
Goals: Research on high performance computing with application to web search and mining.

Projects:
Task-oriented web search: Developed an algorithm to discover user tasks (i.e., group of queries
having the same latent need) from search engine logs. Evaluated F1 score 16% better than tradi-
tional techniques, and about 5% better than the very best method known at that time.
Results published in ACM WSDM 2011 conference (best paper runner up).

SELECTED PUBLICATIONS

Journals
Calzavara, S., Conti, M., Focardi, R., Rabitti, A. and Tolomei, G. Machine Learning for Web
Vulnerability Detection: The Case of Cross-Site Request Forgery. In IEEE Security & Privacy, in
press [impact factor = 1.596].
Tolomei, G. and Silvestri, F. Generating Actionable Interpretations from Ensembles of Decision
Trees. In IEEE Transactions on Knowledge and Data Engineering (TKDE), in press [impact
factor = 3.857].
Tolomei, G., Lalmas, M., Farahat, A., and Haines A. You Must Have Clicked on this Ad by
Mistake! Data-Driven Identification of Accidental Clicks on Mobile Ads with Applications to
Advertiser Cost Discounting and Click-Through Rate Prediction. In International Journal of Data
Science and Analytics, Vol. 7, Issue 1, pp. 53 66.
Aksu, H., Babun, L., Conti, M., Tolomei, G., and Uluagac, A. S. Advertising in the IoT Era:
Vision and Challenges. In IEEE Communications Magazine, Vol. 56, Issue 11, pp. 138 144
[impact factor = 10.435].
Calzavara, S., Tolomei, G., Bugliesi, M., and Orlando, S. A Supervised Learning Approach to
Protect Client Authentication on the Web. In ACM Transactions on the Web (TWEB), Vol. 9,
Issue 3 - June 2015, Article No. 15, pp. 1 30 [impact factor = 1.526].
Giummolè, F., Orlando, S., and Tolomei, G. A Study on Microblog and Search Engine User Be-
haviors: How Twitter Trending Topics Help Predict Google Hot Queries. In ASE Human Journal,
Vol. 2, Issue 3 - September 2013, pp. 195 209.
Lucchese, C., Orlando, S., Perego, R., Silvestri, F., and Tolomei, G. Discovering Tasks from Search
Engine Query Logs. In ACM Transactions on Information Systems (TOIS), Vol. 31, Issue 3 - July
2013, pp. 1 43 [impact factor = 2.312; ACM 2013 Computing Reviews Notable Article]
Miori, V., Tarrini, L., Manca, M., and Tolomei, G. An Open Standard Solution for Domotic
Interoperability. In IEEE Transactions on Consumer Electronics, Vol. 52, Issue 1 - February 2006,
pp. 97 103 [impact factor = 1.694].

Conferences and Workshops
Calzavara, S., Lucchese, C., and Tolomei, G. Adversarial Training of Gradient-Boosted Decision
Trees. In Proc. of ACM CIKM 2019 [to appear] [rank = A].
Calzavara, S., Conti, M., Focardi, R., Rabitti, A., and Tolomei, G. Mitch: A Machine Learning
Approach to the Black-Box Detection of CSRF Vulnerabilities. In Proc. of IEEE Euro S&P 2019,
pp. 528 543.
Conti, M., Gangwal, A., Gochhayat, S. P., and Tolomei, G. Spot the Difference: Your Bucket is
Leaking : A Novel Methodology to Expose A/B Testing Effortlessly. In Proc. of IEEE CNS 2018,
pp. 1 7.
Tolomei, G., Silvestri, F., Haines, A., and Lalmas, M. Interpretable Predictions of Tree-based
Ensembles via Actionable Feature Tweaking. In Proc. of ACM KDD 2017, pp. 465 474 [rank =
A ].
Lucchese, C., Nardini, F. M., Orlando, S., and Tolomei, G. Learning to Rank User Queries to
Detect Search Tasks. In Proc. of ACM ICTIR 2016, pp. 157 166.
Lalmas, M., Lehmann, J., Shaked, G., Silvestri, F., and Tolomei, G. Promoting Positive Post-Click
Experience for In-Stream Yahoo Gemini Users. In Proc. of ACM KDD 2015, pp. 1929 1938 [rank
= A ].
Calzavara, S., Tolomei, G., Bugliesi, M., and Orlando, S. Quite a Mess in My Cookie Jar! Lever-
aging Machine Learning to Protect Web Authentication. In Proc. of WWW 2014, pp. 189 200
[rank = A ].
Giummolè, F., Orlando, S., and Tolomei, G. Trending Topics on Twitter Improve the Prediction
of Google Hot Queries. In Proc. of ASE/IEEE SocialCom 2013, pp. 39 44 [rank = B; among
the top-5% best papers].
Lucchese, C., Orlando, S., Perego, R., Silvestri, F., and Tolomei, G. Modeling and Predicting the
Task-by-Task Behavior of Search Engine Users. In Proc. of OAIR 2013, pp. 77 84.
Orlando, S., Pizzolon, F., and Tolomei, G. SEED: A Framework for Extracting Social Events from
Press News. In Proc. of WWW-WoLE 2013, pp. 1285 1294 [rank = A ].
Ferrari, A., Gnesi, S., and Tolomei, G. Using Clustering to Improve the Structure of Natural
Language Requirements Documents. In Proc. of REFSQ 2013, pp. 34 49 [rank = B; best
paper runner-up].
Bruni, E., Ferrari, A., Seyff, N., and Tolomei, G. Automatic Analysis of Multimodal Requirements:
A Research Preview. In Proc. of REFSQ 2012, pp. 218 224 [rank = B].
Ferrari, A., Gnesi, S., and Tolomei, G. A clustering-based approach for discovering flaws in re-
quirements specifications. In Proc. of ACM SAC 2012, pp. 1043 1050 [rank = B].
Ceccarelli, D., Gordea, S., Lucchese, C., Nardini, F.M., and Tolomei, G. Improving Europeana
Search Experience Using Query Logs. In Proc. of TPDL 2011, pp. 384 395 [rank = B].
Lucchese, C., Orlando, S., Perego, R., Silvestri, F., and Tolomei, G. Identifying Task-based Sessions
in Search Engine Query Logs. In Proc. of ACM WSDM 2011, pp. 277 286 [rank = A ; best
paper runner-up].
Lucchese, C., Orlando, S., Perego, R., Silvestri, F., and Tolomei, G. Detecting Task-based Query
Sessions using Collaborative Knowledge. In Proc. of WI-IAT 2010, pp. 128 131 [rank = B].
Tolomei, G., Orlando, S., and Silvestri, F. Towards a Task-based Search and Recommender Sys-
tems. In Proc. of IEEE ICDE 2010, pp. 333 336 [rank = A ].
Mordacchini, M., Dazzi, P., Tolomei, G., Baraglia, R., Silvestri, F., and Orlando, S. Challenges
in designing an interest-based distributed aggregation of users in P2P systems. In Proc. of IEEE
ICUMT 2009, pp. 1 8.
Tolomei, G. Search the web x.0: mining and recommending web-mediated processes. In Proc. of
ACM RecSys 2009, pp. 417-420 [rank = B].
Miori, V., Tarrini, L., Manca, M., and Tolomei, G . DomoNet: a Framework and a Prototype
for Interoperability of Domotic Middlewares based on XML and Web Services. In Proc. of IEEE
ICCE 2006, pp. 117 118.

TECHNOLOGY TRANSFER
Patent US20170154356A1 (Co-Author)
Title: "Generating actionable suggestions for improving user engagement with online advertisements".
Details: https://patents.google.com/patent/US20170154356A1/en
Patent US20170004542A1 (Co-Author)
Title: "Method and system for providing content supply adjustment".
Details: https://patents.google.com/patent/US20170004542A1/en
Patent US20170004541A1 (Co-Author)
Title: "Method and system for analyzing user behavior associated with web contents".
Details: https://patents.google.com/patent/US20170004541A1/en
Patent US20180247222A1 (Co-Author)
Title: "Changing machine learning classification of digital content".
Details: https://patents.google.com/patent/US20180247222A1/en

Spin-off at Università degli Studi di Padova (Co-founder)
Name: "CAPTCHAd"
Description: CAPTCHAd aims to allow entities which make use of CAPTCHA services (e.g., web por-
tals, blogs, and more generally any service that needs to prevent automatic software bots to access
their resources) to monetize by means of "sponsored challenges", namely CAPTCHA challenges that embed advertising contents.

TEACHING
Big Data Computing [60 hours] 2019 - 2020
M.Sc. in Computer Science (2nd year, 2nd semester)
Sapienza University of Rome, Italy
Operating Systems [60 hours] 2019 - 2020
B.Sc. in Computer Science (2nd year, 1st semester)
Sapienza University of Rome, Italy

Python Programming for Data Science [40 hours] 2018 - 2019
(within Fundamentals of Information Systems)
M.Sc. in Data Science (1st year, 1st semester)
Università degli Studi di Padova, Italy

Introduction to Computer Programming [32 hours] 2018 - 2019
B.Sc. in Computer Science (1st year)
Università degli Studi di Padova, Italy

Python Programming for Data Science [44 hours] 2017 - 2018
(within Fundamentals of Information Systems)
M.Sc. in Data Science (1st year, 1st semester)
Università degli Studi di Padova, Italy

Introduction to Computer Programming [32 hours] 2017 - 2018
B.Sc. in Computer Science (1st year)
Università degli Studi di Padova, Italy

Database [teaching assistant - 25 hours] 2016 - 2017
B.Sc. in Computer Science (2nd year)
Università degli Studi di Padova, Italy

Java Enterprise Edition [60 hours] 01/2014 - 02/2014
Master "SIVE Formazione"
Università Ca' Foscari Venezia, Italy, at SIPE S.r.l.

OTHER EXPERIENCES
Software Engineer 07/2006 - 01/2008
Company: Sysdat Informatica s.r.l., Pisa, Italy
Main responsibilities: Analysis, development, test, and deployment of Java Enterprise (J2EE-compatible)
applications, specifically designed for third-party customers.

Projects:
Supermarket logistics: Developed a web-based software to manage the dispatching of goods for the
logistic department of "Conad" (i.e., one of the largest supermarket chains in Italy). Regularly
interfacing with clients during the whole development stage up to the first product release.
Motorway payment system: Developed the invoicing software system for "Autostrade per l Italia
S.p.A." (i.e., the Italian Concessionaire for toll motorway construction and management handling
about 5 million daily customers, on average).
Software Engineer 12/2005 - 06/2006
Company: NETikos S.p.A., Pisa, Italy
Main responsibilities: Analysis, development, test, and deployment of Java Enterprise (J2EE-compatible)
applications, specifically designed for third-party customers.

Projects:
Mobile network provider web portal: Developed the online recharge secure system for prepaid SIM
cards of private customers inside the web portal of "Telecom Italia Mobile" (i.e., the largest Italian
mobile telecommunications company counting more than 30 million subscribers). Implemented
the electronic shopping cart and the interaction with the electronic payment gateway GestPay,
powered by "Banca Sella S.p.A."

TECHNICAL SKILLS
Programming Languages Python, Java, C/C++, R, Unix scripting (bash, awk, sed, etc.), SQL,
Pig Latin, HiveQL, PHP, JavaScript
Libraries (Python) Keras, Scikit-Learn, Pandas, Numpy, Scipy, Matplotlib, Seaborn
Development Environments JupyterLab, Eclipse, NetBeans
Frameworks Hadoop, Spark
Database MySQL
Other Technologies HTML/CSS, Hadoop, Git, SVN, LATEX

ADMINISTRATIVE STRENGTHS
European Research Projects 01/2008 - 01/2012
Name: FP7 Network of Exellence S-CUBE: Software Services and Systems Network
Unit: ISTI-CNR, Pisa, Italy
Description: Funded by the Commission of European Communities Information Society and Media
Directorate-General, the project mission is to establish a unified, multidisciplinary, vibrant research
community which will enable Europe to lead the software-services revolution and help shape the soft-
ware service based Internet which will underpin the whole of our future society. S-Cube aims to push
the frontiers of research in Service Oriented Computing by creating a vigorous research agenda where
knowledge from diverse research communities is meaningfully synthesized, integrated and applied.
Activity: ISTI-CNR has been responsible for designing novel knowledge extraction techniques from
service-oriented architecture system logs.
Details: http://www.s-cube-network.eu/
Event Organization
Executive Director of the "2019 International Summer School on Machine Learning and Security"
Details: 2019 ML&S School
Keynote Speaker at the "Yahoo Tech Pulse 2016" Conference
Title: "A Taste of Machine Learning"
General Program Chair of the "IEEE Security and Privacy in Digital Advertising Workshop" (IEEE
CNS SPA 2017)
Details: IEEE CNS SPA 2017
Program Co-Chair of the "IEEE Cyber-Physical Systems Security Workshop" (IEEE CNS CPS-Sec
2018)
Details: IEEE CNS CPS-Sec 2018
Program Co-Chair of the "IEEE Cyber-Physical Systems Security Workshop" (IEEE CNS CPS-Sec
2017)
Details: IEEE CNS CPS-Sec 2017
Program Committee Membership

Conferences ACM WSDM 2017-2020, ACM CIKM 2017-2019, ACM KDD 2019, ACM SIGIR 2018-
2020, ACM ASONAM 2015-2020, ACM ICTIR 2017-2018, ECML/PKDD 2018-2019,
TheWebConf (former WWW) 2018-2020, IJCAI 2016
Workshops IIR 2015-2018