EE7001/EE8001: CS Seminar, 114-1, Fall 2025

Course Staffs

• Instructor: Shih-Han Hung
• TA: George Cheng

Schedule and Grading

• Location: MD-231 (明達231)
• Total seminars: 12 sessions (Week 2-15, no seminar on September 29 and October 6).
• You must earn at least 70 points to pass the course.
• Attendance: 7 points per seminar
• Asking one question during a seminar: 2 points

Absence Policy

• Absences may be excused with valid proof.
• Please notify the TA via NTU COOL or email if you cannot attend a seminar.

Student Speaker

• Students may volunteer to give a seminar talk.
• Student speakers are exempt from attending all the other seminars.
• Please email the Instructor if you are willing to give a talk by September 30.
• If a scheduled student talk is cancelled for any reason, the speaker will record and submit a video presentation.

Schedule (Tentative)

Week 2 (September 8)

• Speaker: William Wang (UC Santa Barbara)

• Title: How Agentic AI is Reinventing Chip Design and Verification

• Abstract: Semiconductor innovation is at a critical juncture, demanding next-generational methods to overcome rising complexities, shorter design cycles, and intense competitive pressures. Traditional EDA tools are constrained by manual processes and limited intelligence, but what if we could transcend these limitations? Enter AI Agents, the AI solution leveraging large language models and advanced algorithms to continue to improve themselves. In this talk, Prof. William Wang, Founder & CEO of ChipAgents, will introduce how AI agents go beyond traditional EDA automation, embedding agentic intelligence capable of independently handling hardware modeling, constraint-solving, automated debugging, testbench generation, and even proactive design optimization. Highlights include Use Cases, Scalability & Reliability: Case studies illustrating substantial productivity improvements, enhanced design quality, and accelerated time-to-market achieved by leading semiconductor enterprises deploying AI Agents. AI Agents in Action: Real-world scenarios demonstrating how AI agents autonomously identify critical bugs, optimize RTL designs, and significantly shorten verification cycles.

• Bio: William Wang is an internationally recognized pioneer in artificial intelligence and the visionary Founder and CEO behind ChipAgents.ai. He’s also Duncan and Suzanne Mellichamp Endowed Chair Professor in AI and Designs at UC Santa Barbara. His pioneering research has earned numerous prestigious accolades, including the IEEE SPS Laplace Award, NSF CAREER Award, DARPA Young Faculty Award, the British Computer Society - Karen Sparck Jones Award, and IEEE AI’s 10 to Watch. William’s vision drives ChipAgents’ mission to transform semiconductor design through intelligent, agentic AI systems.

Week 3 (September 15)

• Speaker: Hsin-Po Wang (National Taiwan University)

• Title: Ambidextrous Degree Sequence Bounds for Pessimistic Cardinality Estimation

• Abstract: In a large database system, upper-bounding the cardinality of a join query is a crucial task called pessimistic cardinality estimation. Recently, Abo Khamis, Nakos, Olteanu, and Suciu unified related works into the following dexterous framework. Step 1: Let $(X_1, …, X_n)$ be a random row of the join, equating $H(X_1, …, X_n)$ to the log of the join cardinality. Step 2: Upper-bound $H(X_1, …, X_n)$ using Shannon-type inequalities such as $H(X, Y, Z) \leq H(X) + H(Y|X) + H(Z|Y)$. Step 3: Upper-bound $H(X_i) + p H(X_j | X_i)$ using the $p$-norm of the degree sequence of the underlying graph of a relation. While old bound in step 3 count claws in the underlying graph, we proposed ambidextrous bounds that count claw pairs. The new bounds are provably not looser and empirically tighter: they overestimate by $x^{3/4}$ times when the old bounds overestimate by $x$ times. An example is counting friend triples in the com-Youtube dataset, the best dexterous bound is $1.2 \times 10^9$, the best ambidextrous bound is $5.1\times 10^8$, and the actual cardinality is $1.8 \times 10^7$.

• Bio: Hsin-Po Wang is an Assistant Professor in EE and GICE at NTU. His research interests lie in information theory and coding theory, where he applies techniques in algebra, combinatorics, and probability theory to polar codes, group testing, distributed storage, distributed computation, sampling algorithms, and database optimization. Hsin-Po earned his B.Sc. in Math at NTU and completed his Ph.D. in Math at UIUC. He has held research positions at UC San Diego, UC Berkeley, and the Simons Institute for the Theory of Computing. Known for his extensive, creative use of Tikz figures in papers, Hsin-Po is equally passionate about speedrun techniques for educational purposes and assembling binder clips into fullerene-like structures.

Week 4 (September 22)

• Speaker: Yen-Huan Li (National Taiwan University)

• Title: From data compression to learning quantum states

• Abstract: This talk will cover three closely related problems: universal coding for data compression, online portfolio selection for long-term investment, and online learning of quantum states for quantum state tomography. While the application scenarios appear distinct, these problems indeed share similar mathematical structures. I will offer a brief introduction to all three problems and delve into state-of-the-art results for the latter two.

• Bio: Yen-Huan Li received the B.S. degree in electrical engineering and the M.S. degree in communication engineering from National Taiwan University in 2008 and 2010, respectively, and the PhD degree in computer science from École polytechnique fédérale de Lausanne (EPFL) in 2018. From 2018 to 2023, he was an assistant professor in the Department of Computer Science and Information Engineering at National Taiwan University. Since 2023, he has been an associate professor in the same department. He is interested in developing rigorous algorithms for decision making under uncertainty with as weak assumptions as possible. Professor Li received the Young Scholar Fellowship from the Ministry of Science and Technology of Taiwan and the Young Theoretical Scientist Award from the National Center for Theoretical Sciences of Taiwan.

Week 5 (September 29): no seminar, Teachers’ Day

Week 6 (October 6): no seminar, Mid-Autumn Festival

Week 7 (October 13)

• Speaker: Tsung-Shou Liao (National Taiwan University)

• Title: From online algorithms to learning-augmented online algorithms

• Abstract: In the past three decades, online algorithms have gained a considerable amount of research interest. The concept of online algorithms has been actually applied to a variety of computational models with uncertainty (e.g., online learning), while the only theoretical measure to evaluate the performance of an online algorithm is “competitive ratio”; that is, comparing an online algorithm to an offline optimal adversary. The fundamental question in the field of online algorithms is that: Can we evaluate the quality of an online algorithm in a better way? A new concept, called learning-augmented algorithms, or algorithms with predictions, was proposed to answer the question in 2018. Precisely, the notion exploits the prediction power of popular machine learning models to obtain the robustness guarantees of competitive analysis for online algorithms. In this talk, we first introduce several classical problems as well as their online algorithms. Next, we discuss not only how to devise online algorithms with learning predictions in order to obtain a better theoretical guarantee, but also how to provide a way to look into different performance evaluation beyond the most common worst-case analysis.

• Bio: Chung-Shou Liao is a professor in Dept Electrical Engineering at National Taiwan University (NTU). Before joining NTU, he was a distinguished professor in Dept Industrial Engineering at National Tsing Hua University (NTHU), where he had worked for 15 years. At the beginning of his career, he conducted research on graph theory and geometric computing at the Institute of Information Science, Academia Sinica, while pursuing his Ph.D. in Dept Computer Science at NTU.

  Dr. Liao’s research focuses on designing efficient algorithms for solving combinatorial optimization problems from real-world applications. He received the NSTC Outstanding Research Award in 2023 and the Ta-You Wu Memorial Award in 2016. In 2019, he was a Fulbright Senior Research Scholar at CSAIL, MIT.

  Dr. Liao served as the PC Chair for AAAC 2021 (the 14th Annual Meeting of the Asian Association for Algorithms and Computation) and ISAAC 2018 (the 29th International Symposium on Algorithms and Computation). He is a board member of AAAC, an advisory committee member of ISAAC, and an Associate Editor of Journal of Combinatorial Optimization and International Journal of Foundations of Computer Science. In recent years, his lab has collaborated with high-tech companies such as MediaTek, UMC, TSMC, and Unimicron in the semiconductor manufacturing industry.

Week 8 (October 20)

• Speaker: Yu-Fang Chen (Academia Sinica)

• Title: An Automata-Based Approach for Quantum Program Verification

• Abstract: Ever wondered how we can automatically verify quantum programs? In this talk, I will present a framework that does exactly that. Our approach models quantum states as decision trees and sets of states as tree automata, enabling precise yet scalable reasoning about program behavior.

  We verify properties of the form {P} C {Q}, where P and Q are pre- and post-conditions modeled as tree automata, and C is the quantum program. Intuitively, we ensure that every state in P ends up in some state in Q after executing C. Each quantum gate is equipped with a transformer that updates the automaton to reflect the gate’s effect. By composing these transformers, we simulate all reachable states and verify correctness via language inclusion.

  I will show how this framework naturally supports tasks like circuit equivalence, and how extending tree automata with quantum-specific features makes the approach even more powerful. Throughout the talk, I will interleave live tool demonstrations to demonstrate the verification process on a series of quantum examples. The talk is based on a series of works published at PLDI’23 (Distinguished Paper, CACM Research Highlights), CAV’23, TACAS’25, and POPL’25.

• Bio: Yu-Fang Chen is a Research Fellow at the Institute of Information Science, Academia Sinica, Taiwan. His research spans automata theory, program verification, and programming languages, with pioneering contributions in two main directions: automated verification of quantum programs and satisfiability modulo theories on strings (SMT on strings).

  In quantum verification, he developed the AutoQ framework, the first automata-based system capable of fully automated verification of large-scale quantum circuits. He introduced models such as symbolic amplitudes and level-synchronized tree automata, which enable the verification of properties that were previously beyond reach. These results have appeared at PLDI, POPL, CAV, and TACAS, earning multiple Distinguished/Best Paper Awards (PLDI 2023, OOPSLA 2023, FM 2023, TACAS 2010). His work was featured as a CACM Research Highlight and internationally recognized, including a keynote talk at Highlights 2025.

  He regularly serves on program committees of leading conferences such as CAV, TACAS, LICS, and PLDI, and is the organizer of Dagstuhl Seminar 26111 on Formal Analysis and Verification in Quantum Programming Languages and co-chair of the VQC 2025 workshop at CAV.

Week 9 (October 27): student talk

• Speaker: Chih-Kai Yang (National Taiwan University)

• Title: AudioLens: A Closer Look at Auditory Attribute Perception of Large Audio-Language Models

• Abstract: Understanding the internal mechanisms of large audio-language models (LALMs) is crucial for interpreting their behavior and improving performance. This work presents the first in-depth analysis of how LALMs internally perceive and recognize auditory attributes. By applying vocabulary projection on three state-of-the-art LALMs, we track how attribute information evolves across layers and token positions. We find that attribute information generally decreases with layer depth when recognition fails, and that resolving attributes at earlier layers correlates with better accuracy. Moreover, LALMs heavily rely on querying auditory inputs for predicting attributes instead of aggregating necessary information in hidden states at attribute-mentioning positions. Based on our findings, we demonstrate a method to enhance LALMs. Our results offer insights into auditory attribute processing, paving the way for future improvements.

• Bio: Chih-Kai Yang is a second‐year M.S. student at National Taiwan University, advised by Prof. Hung‐yi Lee. His research interests cover speech processing and natural language processing (NLP), with a particular focus on the analysis and the development of speech processing models, large language models (LLMs), and their interaction and integration in research and applications. He is also deeply interested in speech large language models (speech LLMs), which facilitate more natural human‐AI conversations through speech.

Week 10 (November 3)

• Speaker: Ya-Ting Yang (New York University)

• Title: Design for Security and Resilience in AI-Driven Cyber Physical Human Systems

• Abstract: Modern societies increasingly rely on AI-driven cyber-physical-human systems (CPHSs), such as intelligent transportation, industrial automation, and other critical infrastructure. While these systems promise efficiency and intelligence, they also introduce new vulnerabilities in which security and resilience are tightly coupled with trust between system components. This raises a fundamental question: how can we design socio-technical systems that remain trustworthy and resilient in the presence of adversarial manipulation and the cognitive biases inherent in human decision-making? In this talk, we will present a research agenda aimed at developing principled yet computationally tractable frameworks for understanding and engineering trust in CPHSs, drawing on insights from game theory and cognitive science. We will walk through four complementary perspectives: assessing, establishing, maintaining, and exploiting trust, with a particular emphasis on how adversarial trust and belief can be strategically exploited through cross-layer defensive deception design. The talk will conclude by outlining future directions toward resilient, cognitive-aware CPHSs.

• Bio: Ya-Ting Yang is a Ph.D. candidate in Electrical and Computer Engineering at New York University, affiliated with the NYU Center for Cybersecurity. She received her M.S. in Communication Engineering from National Taiwan University and her B.S. in Electrical Engineering from National Tsing Hua University. Her research focuses on game theory and optimization with applications to the security and resilience of AI-driven cyber-physical-human systems. She is an RSAC Security Scholar, and her contributions have been published in leading journals and conferences, including IEEE TIFS, TNSE, TNSM, TITS, IOTJ, CDC, and Globecom.

Special Seminar I (November 6)

• Speaker: Vincent Hwang (Max Planck Institute)

• Title: Cryptographic Engineering in Post-Quantum Cryptography

• Abstract: Cryptography engineering studies how cryptosystems could/should be best implemented in real-world computing devices. We want cryptosystems to be fast, secure, and correct. This talks goes over two published works the speaker authored/coauthored in this line of research. The first part of this talk studies the impact of the constant-time programming principle while optimizing the National Institute of Standards and Technology (NIST) post-quantum cryptography standard FIPS204 (ML-DSA). On the processor Cortex-M3, long multiplication instructions have input-dependent execution time, and instructions terminate earlier if the results have absolute values smaller than a certain threshold. A straightforward workaround is to emulate the long multiplication with multi-limb arithmetic. We generalized the Barrett-type modular multiplication and demonstrated its efficiency for multi-limb arithmetic. The second part of this talk studies the formal verification of the software-emulated floating-point arithmetic in the submission package of the Falcon, an upcoming NIST PQC standard FIPS206 (FN-DSA). In the submission package of Falcon, the authors implemented the floating-point arithmetic with software emulation to ensure constant-time computations. The speaker (i) found a discrepancy between the emulated floating-point multiplications and the claimed behavior; (ii) verified the absence of the discrepancy in the Falcon implementations with respect to a model in the domain-specific language CryptoLine; and (iii) demonstrated the equivalences between software-emulated floating-point arithmetic.

• Bio: Vincent Hwang is a third-year PhD student at Max Planck Institute for Security and Privacy, Bochum, Germany under the supervision of Peter Schwabe. He obtained a bachelor’s (2021) and a master’s (2022) from the Department of Computer Science and Information Engineering at National Taiwan University, Taipei, Taiwan. His main research focuses are assembly optimization in cryptographic engineering on various platforms, formal verification for assembly-optimized programs, and recently elliptic-curve discrete logarithms on data-center-level GPUs. He authored/coauthored two publications in his bachelor’s studies, six publications in his master’s studies, and seven publications in his PhD studies. He is enthusiastic on turning high-level constructs in cryptography into highly-optimized assembly programs.

Week 11 (November 10)

• Speaker: Tsun Ming Cheung (Academia Sinica)

• Title: Communication Complexity as an Interface to Theoretical Computer Science and Applications

• Abstract: In the basic communication complexity model, two or more players collaborate to determine the value of a joint function, where each player privately holds a portion of the input. The central goal of communication complexity theory is to quantify the minimal amount of information that must be exchanged to compute different functions, under the assumption that each player has unbounded computational power. Though a seemingly unrealistic model, communication complexity has proven highly useful for establishing impossibility results across many areas of computer science, both theoretical and applied.

  In this talk, I will survey several results from my past research in communication complexity, highlighting applications in other areas like algorithm design and query complexity theory. I will illustrate how communication complexity techniques yield provable hardness results, providing insights into the fundamental limits of computational tasks and guiding the design of efficient algorithms.

• Bio: Tsun-Ming Cheung is a postdoctoral fellow at the Institute of Information Science, Academia Sinica. His research interests span theoretical computer science and discrete mathematics, encompassing areas such as complexity theory, streaming algorithms, and quantum computation theory. Previously, he received his PhD degree from McGill University, Canada, and his M.Phil. degree from the Chinese University of Hong Kong.

Special Seminar II (November 11)

• Speaker: Albert Wu (Stanford University)

• Title: Unifying Physics and Learning for Generalizable Robotic Manipulation

• Abstract: Albert Wu (吳孟忻) is a Ph.D. candidate in Computer Science at Stanford University. He received his B.S. and M.Eng. in Electrical Engineering and Computer Science from MIT. His research lies in robotic manipulation and hybrid-system motion planning, with a focus on exploiting underlying mathematical structures to achieve generalizable and robust manipulation. Drawing on tools from optimization, control theory, and differential geometry, he develops unified approaches that bridge physics-based models and learning algorithms. Albert’s research has been validated on a range of real-world systems, including dexterous grasping, long-horizon extrinsic manipulation, and bimanual manipulation.

• Bio: Robotic manipulation remains one of the most fundamental yet unsolved challenges in robotics. Unlike navigation or locomotion, manipulation involves non-smooth, aperiodic, and richly constrained hybrid dynamics that make it difficult to model and control. In this talk, I will show how exploiting the underlying physics and structure of manipulation dynamics leads to more robust, interpretable, and generalizable policies. First, I will show how contact conditions provide a natural representation for organizing hybrid modes and composing manipulation skills. Second, I will introduce differential-geometric tools that capture how high-degree-of-freedom robots move on low-dimensional manifolds induced by kinematic and contact constraints. Finally, I will discuss how these representations can be integrated with control-theoretic formulations to achieve safety and verifiability in real robotic systems. Using real-world examples from dexterous and extrinsic manipulation, I will show how these components can form the foundation of a unified framework that combines physics-based models and learning to enable robots that can reason, act, and adapt across tasks and embodiments.

Week 12 (November 17): student talk

• Speaker: Bain Chang (National Taiwan University)

• Title: AI提問應用及對市場、勞動力影響

• Abstract: AI (Artificial Intelligence) has become an indispensable part of current technological and societal development. This speech will elaborate on AI from two main perspectives.The first is to help you “talk to AI” in a way that reliably produces useful results—by mastering a handful of prompt‑engineering habits that anyone can adopt. The second is to make sense of how generative AI is reshaping markets and work, with a grounded view of Taiwan’s strengths in the AI server supply chain and concrete deployment paths—from cloud to on‑prem to single‑machine setups. The intent is that you leave with methods you can use tomorrow and a clear mental model of where value—and risk—will likely concentrate. Ultimately, the goal is to empower you to engage with AI not just as a user, but as a strategic thinker—someone who can harness its capabilities responsibly and effectively in both professional and societal contexts.

• Bio: Bain Chang is a Ph.D. Student in Electronic Engineering at NTU. He received his B.S. in Information Management and M.Eng. in Electrical Engineering from NTU. He previously worked at Yahoo and Trend Micro, and also served as a university lecturer. He is currently working at the Central Bank (or Central Bank of Republic of China(Taiwan)), primarily focusing on FinTech-related research. His main responsibilities include:, System Test Automation Design, Machine and Deep Learning Model Development/Building Research on Various Financial Indicators (e.g., interest rates, housing prices, commodity prices, etc.) and LLM Architecture Development/Building (or Construction), Deployment, and Refinement of Prompt Engineering Techniques.

Week 13 (November 24)

• Speaker: Hung-Wen Chen (National Tsing Hua University)

• Title: Physics-Guided Photonic AI: From Laser Color Prediction to TFT-LCD Defect Classification

• Abstract: This talk covers two independent applications of AI-driven photonic technologies. First, we present a physics-driven tandem neural network (PD-TNN) that addresses the inverse problem of ultrafast laser coloration. By incorporating two physics-derived features—average fluence (F) and average time per unit area (τ)—together with a perceptual ΔE loss, PD-TNN mitigates the one-to-many mapping inherent in laser-induced thin films and enables highly predictable, and perceptually accurate metal color generation for advanced manufacturing. Second, we introduce a multimodal automatic defect classification (ADC) system for TFT-LCD panels based on Descriptive Embedding Generation (DEG). The method provides strong generalization under zero-shot and few-shot conditions, improving defect recognition accuracy while reducing the need for manual inspection.

• Bio: Hung-Wen Chen received his Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology and is now an Associate Professor in the iPhD Program at National Tsing Hua University, jointly appointed with the Institute of Photonics Technologies and the Department of Industrial Engineering and Engineering Management. His research focuses on ultrafast-laser photonics and AI-driven modeling for precision laser machining and smart manufacturing. Dr. Chen has received multiple national distinctions, including the Future Tech Award, the National Innovation Award, and the 2023 Delta Young Scholar Award in Smart Manufacturing. He is also the founder of m’AI Touch Technology Co., Ltd., dedicated to translating photonic–AI innovations into real-world intelligent systems.

Week 14 (December 1)

• Speaker: Siyao Guo (NYU Shanghai)

• Title: No Time to Hash: On Super Efficient Entropy Accumulation

• Abstract: Real-world random number generators (RNGs) cannot afford to use (slow) cryptographic hashing every time they refresh their state R with a new entropic input X. Instead, they use “superefficient” simple entropy-accumulation procedures, such as R ← rot_{α,n}(R) ⊕ X, where rot_{α,n} rotates an n-bit state R by some fixed number α.For example, Microsoft’s RNG uses α = 5 for n = 32 and α = 19 for n = 64. Where do these numbers come from? Are they good choices? Should rotation be replaced by a better permutation π of the input bits?In this talk, we aim to provide a rigorous study of these pragmatic questions.Joint work with Yevgeni Dodis (NYU), Noah Stephens-Davidowitz (Cornell University) and Zhiye Xie (NYU Shanghai).Paper link: https://eprint.iacr.org/2021/523.pdf

• Bio: Siyao Guo is an associate professor of Computer Science at NYU Shanghai. Her research interests are foundations of cryptography, computational complexity and pseudorandomness.

Week 15 (December 8): no seminar

Week 16 (December 15): no seminar, final exam