Marc Light Email and Phone Number

I am part of ADI's AI Group which develops AI solutions that aim to delight both ADI's customers and its business stakeholders. More specifically, we work with ADI business verticals to infuse ADI products with AI, making the Intelligent Edge a reality. We also work ADI engineers, marketers, sales, legal, and HR teams to improve internal operations by leveraging AI technologies. Impact- Initiated a trio of genAI proof-of-concept projects (loosely based on the ChipNemo paper https://arxiv.org/abs/2311.00176 ) collaborating with stakeholders (eg., Tom Roseanu) to gather requirements, started their implementation, and transitioning them to newly hired team members, and finally, worked with ADI IT to provide a cloud environment to develop, deploy, and demo all three projects in a secure way. - Implemented an image embedding algorithm for an internal image search system that is used by ADI engineers.- Served as the AI team’s technical lead on the deployment of the SambaNova Suite and its pilot use for assisting ADI Field Application Engineers in product selection.- Assisted partners with the software toolchain for the ADI MAX7800: an AI Microcontroller and CNN accelerator and helped determine its suitability with respect to latency and power consumption.- Hired four AI Group team members, developed an interview sequence, worked with the recruiting team to develop a pipeline, performed the interviews, and trained others to perform the interviews. Our hiring is ongoing.- Managed an intern from University of Illinois at Urbana-Champaign ECE working on an edge AI application for low-latency denoising: https://www.linkedin.com/posts/erkan-bayram-uiuc_i-have-successfully-completed-my-internship-activity-7232813279587311616-te07 - Collaborated with ADI software and hardware engineers to characterize and simulate AI workloads relevant to ADI customers

I supported a team of 3 cybersecurity researchers, which conducted cybersecurity research utilizing Censys scan data. The research demonstrated thought leadership, built brand, and generated inbound leads and thus we worked closely with our Marketing org. In addition, we assisted the Federal Sales team with POV reports. We also worked with Engineering to improve Censys’ scanning capabilities and I worked with Product to infuse our SaaS offering and underlying data with DS/ML features.

I supported a 10 person team in the Boston area. We were part of the Infrastructure Data Science (IDS) organization which strived to enable data-driven decisions for Facebook infrastructure, helping make it more reliable, efficient, and performant. The Boston team worked in 7 areas ranging from data warehousing to wireless communications and delivered 20+ projects each half. Impact- Built a team of 5 data scientists working with Core Data, middleware that supports low latency data access for the family of Facebook apps. Core Data leadership recognized the team’s impact and transferred two of their SWE hiring lines to expand our team.- Managed team of 4 data scientists working with Messaging Infra to ensure that Facebook Messenger remained reliable during COVID lockdowns, data center shortages, New Year’s Eve, etc.; we utilized numerical optimization techniques.- Helped build IDS from 300 to 400+ data science and data engineers by interviewing 1-2 candidates a week and shepherding them through the hiring process. - Facilitated the continued high impact of the 10 individual contributors in Boston, which is the primary responsibility of Facebook managers (see my Medium post on the topic which is linked in the Publications section below).

I served as supervisor to a team of data ops, data science, and cyber security research folks located in Boston, Raleigh, and Lisbon (not respectively). The team numbered twenty strong and strived to stay aligned with the needs of our customers and demands of our marketplace (which BitSight created). It worked from ideation through product release and beyond to sales, support, & success of the product. Members of the team collaborated with Product Management, Engineering, Customer Success, Customer Support, Marketing, and Sales. Keeping up with current research and trends can be difficult in a startup environment but they did a number of reading groups, hackathons, conferences, and meetups in an attempt to avoid tunnel vision. Impact- Expanded all three teams with a number of key hires. - Facilitated culture of learning and collaboration by organizing hackathons, quarterly lightning talks, and reading groups- Built reputation of teams internally, resulting in many team member contributions to Product, Marketing, and Sales. - Met with customers and worked on our product strategy. I also did some data science in the form of iPython/Pandas/matplotlib code, mostly to satisfy customer-requested analyses. - Supported Sales and Marketing teams by presenting in Switzerland and German (in German) and assisting at a trade conference and speaking as a radio guest in Singapore. - Worked with patent lawyers to build our BitSight’s patent portfolio contributing directly to five cybersecurity patents

I served as a player/coach on a six-person Data Science Team. We strived to keep BitSight Ratings the best in the business, enable new features and products, provide thought leadership in the cyber security industry, and improve other BitSight business functions where possible by infusing them with data science techniques.Impact- Doubled the team size with three additional hires- Met with customers, identified their pain points, ideated solutions, coded proof-of-concept implementations, and worked with Product and Engineering to bring these features to our customers. - Peer analytics allowing customers to understand how they measure up to their peers on ratings and on relevant risk vectors - Forecasting of BitSight Ratings change based on “what if” scenarios - Fourth-party risk modeling - Third-party risk alert threshold setting assistant, to avoid alert fatigue.

I led a growing team of data science / machine learning / operations research people who worked on projects across Honeywell's four divisions: Aerospace, Performance Materials & Technologies, Home & Building Technology, and Safety & Productivity Solutions. The team was located in Atlanta, Minneapolis, Morris Plains, and Bangalore. Our focus was on new analytics-driven products. Impact- Built a foundation of analytics use cases- Helped educate company leadership on what problems analytics, ML, AI, etc., could and could not deliver given Honeywell’s data

I worked to accelerate a move into data-enabled products at Honeywell. I served as a supervisor for a team with expertise in machine learning, control theory, NLP, physics-based modeling, ontologies, video processing, and wearable computing. We utilized Hadoop, Spark, and Cassandra clusters, working with data produced by Honeywell devices which customers use in connected homes, buildings, and industry. We collaborated with Honeywell business units and with members of the labs in Bangalore, Beijing, Bucharest, Prague, and Shanghai. I also assisted product managers with voice-of-the-customer interviews and observations.

Assembled and managed a team of six with advanced degrees in statistics, electrical engineering, physics, and computer science. The group designed and implemented systems for forecasting electrical system load, control center workload, wind power levels, transmission congestion, and hurricane damage. Impact- We built systems for detecting consumer theft and predicting feeder outages. These systems are all being used by our affiliate companies: Florida Power & Light and NextEra Energy Resources. - In addition to managing the group’s activities, in 2013 I focused on the Texas wholesale energy market, modeling price and designing trading strategies that optimize revenue while maintaining an acceptable risk profile. A strategy I helped design is being used for NextEra’s wind farms in Texas. - I also build systems that identify outlier utility bills and faulty electricity meters based on usage, voltage, event, and customer data. These systems are implemented in R and Java and make use of Random Forests, Drools, HBase, and MapReduce and are in production at FP&L. - As part of WindLogics leadership team, I was involved in strategic decision making with respect to both business and technology.

Designed and built systems for finding names in text, using conditional random field statistical models. Modified the prediction algorithm to incorporate heuristics, trade off false-positive rate against false-negative rate, and provide local confidence estimates. Implemented global confidence scores using a logistic regression. Worked with product developers to leverage the system in four products; was involved from ideation through release. I was also part of a team that designed and implemented of a legal search tool that uses a ranking support vector machine to integrate signals from a case's content and previous citations to that case.

Joint appointment in the School of Library & Information Science and in the Linguistics Department with a courtesy appointment in the Computer Science Department.Developed and applied human language processing technology to bioscience literature, employing conditional log-linear models, hidden Markov models, and support vector machines. Developed systems for automatically populating a database of transcription factor interactions, locating scientific speculations in abstracts, highlighting scientific articles, and indexing Pubmed abstracts by named entities such as proteins, genes, and organisms. Received a three year contract extension in spring 2005.

Led development of a system capable of extracting from a corpus of newswire text specific answers to user questions (rather than a list of relevant documents). Led two related projects, managing a team of ten people. Secured funding totaling over two million dollars from internal and external sources.

Led the construction of a shallow-but-broad-coverage context-free grammar for German and trained a corresponding head-lexicalized probabilistic context-free grammar. Trained probabilistic model on raw newspaper text using an EM algorithm.

Designed and implemented a system that learns information about words from syntactically annotated corpora, utilizing hidden Markov models trained using an EM algorithm. Worked with international team of researchers to design and build a speech-to-speech translation system called Verbmobil. Built a module to assign shallow syntactic structure to spoken utterances, using finite-state transducers and hidden Markov models.