Meet the NEC Labs America Intern Helping to Make Autonomous Vehicles Safer and More Secure

September 8, 2023 | NEC Labs America | News | Tags: interns, internships, autonomous-vehicles

There’s much more to autonomous vehicle security than locking a car door.

This summer, Kaiyuan Zhang, a 3rd-year computer science Ph.D. student at Purdue University, joined NEC Labs America’s popular intern program to help advance research around autonomous vehicle security.

Each year, nearly 50 Ph.D. candidates join NEC Labs America’s innovative program, which centers on a collaborative environment where interns work directly with senior researchers and potential end-user customers. They focus on advancing cutting-edge technologies at the intersection of machine learning and science that will improve how the world can become more responsive to the needs of individuals. This 12-week program is run out of NEC Labs America’s research facilities in Princeton, NJ, and San Jose, California.

Kaiyuan Zhang NEC Labs America Intern

In addition to an immediate supervisor, each intern is assigned to a specific mentor. Kaiyuan’s mentor this summer is Francesco Pittaluga, a researcher in the Media Analytics Department at NEC Labs America focused on Computer Vision, Machine Learning, and Computational Photography Biography.

According to Pittaluga, “The intern program is not only an excellent opportunity for Ph.D. students to gain experience working in an industrial research lab, but it is also a huge asset for NEC Labs. Interns like Kaiyuan help introduce new insights and ways of thinking to our group. His background in security has been critical for developing an adversarial robustness framework for training and evaluating autonomous driving systems.”

Interns like Kaiyuan quickly become part of a project team applying innovative technology to industry-leading concepts.

“My research interests mainly focus on security and privacy in machine learning,” said Kaiyuan. “With this expertise, I’m able to provide unique perspectives in our collaborative work with other researchers on NEC Labs America’s Media and Analytics team working on improving the robustness of the autonomous driving system.”

This work extends well beyond the physical security of autonomous vehicles. As with most emerging technologies, securing the data generated, processed, and used by autonomous cars is complex. Kaiyuan is helping to develop safety-critical driving scenarios that improve the robustness and safety of autonomous driving systems.

Take, for example, the multitude of information captured by the cameras and sensors equipped on autonomous vehicles. They record a vast array of data from the surrounding environment, everything from other vehicles and traffic signals to hazard cones and pedestrians, even down to lane markings. Kaiyuan is actively involved in automatically generating safety-critical driving scenarios, which is crucial in enhancing the real-world safety of autonomous driving.

Autonomous Vehicle

According to Kaiyuan, there have already been instances of security breaches by malicious entities that manipulate how a car is programmed to respond in specific situations. This includes programming a vehicle to exhibit unreasonable behavior upon detecting orange hazard cones. In another case, adversaries have strategically placed adversarial patterns on stop signs, compelling the vehicle to take an action contrary to stopping. These scenarios carry not just potential danger but could also escalate to lethal circumstances. Research conducted by Kaiyuan and his team is meticulously targeted toward enhancing the robustness of the autonomous driving system against such security threats.

Kaiyuan also hopes to return to the NEC Labs Intern program next summer to continue collaborating on cutting-edge technologies.

Please click here to learn about our NEC Labs America Intern Program.

Read Our News Posts

Mix-CLAP Teaching Audio AI to Work in the Noisy Real World

Mix-CLAP: Teaching Audio AI to Work in the Noisy Real World

Mix-CLAP from NEC Laboratories America delivers near-Transformer accuracy for sound event classification at a fraction of the compute cost, using dual lightweight encoders and adaptive, noise-aware inference for real-world edge deployment.
Read more
How AI Can Transform the Way Companies Buy What They Need

How AI Can Transform the Way Companies Buy What They Need

Procurement teams lose time and money to inaccurate demand forecasts and manual supplier negotiations. A new framework from NEC Corporation and NEC Laboratories America combines automated negotiation with multimodal AI forecasting to optimize both sides of the procurement process.
Read more
Open SAT How We Taught AI to Search Satellite Images Like a Search Engine

Open SAT: How We Taught AI to Search Satellite Images Like a Search Engine

Satellite imagery is vast, high-resolution, and rich with information, but finding specific objects within it using natural language has remained a stubborn challenge. Open-SAT, developed by researchers at NEC Laboratories America and North South University, tackles this problem without retraining any models.
Read more
Training Small AI Models Without Blindly Trusting Big Teacher Models

Training Small AI Models Without Blindly Trusting Big Teacher Models

Machine learning is shifting from learning from data alone to learning from both data and teacher models. Beta-KD uses uncertainty-aware Bayesian weighting to train compact multimodal AI without blindly trusting every teacher signal.
Read more
How Rule-Driven Routing Makes Retrieval-Augmented Generation Smarter

How Rule-Driven Routing Makes Retrieval-Augmented Generation Smarter

Most retrieval-augmented generation systems stop at documents, ignoring the relational databases that power finance, healthcare, and research. Our researchers built a rule-driven framework that learns which source to query for each question, delivering better answers at lower computational cost.
Read more
Making Video AI Fast Enough for the Real World

Making Video AI Fast Enough for the Real World

State-of-the-art video models are accurate but too slow for live deployment. This work transfers their knowledge into causal streaming models that process video frames in real time, achieving 4x lower latency with competitive accuracy across action detection and pedestrian intent tasks.
Read more