Select a calendar:
Filter December Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for December
-
PhD Thesis Proposal - Zunchen Huang
Fri, Dec 02, 2022 @ 09:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Zunchen Huang
Title: Constraint Based Analysis for Persistent Memory Programs
Time: Friday, December 2, 9:00 AM-11:00 AM PST
Committee: Chao Wang (chair), William GJ Halfond, Mukund Raghothaman, Srivatsan Ravi, and Pierluigi Nuzzo.
Abstract: Emerging persistent memory (PM) technologies are beginning to bridge the gap between volatile memory and non-volatile storage in computer systems, by allowing high-speed memory access, byte-addressability, and persistency at the same time. However, PM programming remains a challenging and error-prone task due to reliance on ordinary developers to write correct and efficient PM software code. In this presentation, I propose a framework to detect and repair PM bugs in software code automatically using a set of new symbolic analysis techniques. Unlike existing methods, which rely on patterns and heuristics to detect and repair a small subset of PM bugs, the proposed symbolic analysis framework is able to handle a wide range of PM bugs uniformly. This is achieved by first encoding the program semantics, correctness properties, and PM requirements as a set of logical constraints, and then solving these constraints using off-the-shelf solvers. By reasoning about these logical constraints symbolically, our method can detect, diagnose, and repair PM bugs both efficiently and automatically. It can also infer PM requirements automatically. Finally, I will discuss potential extensions of the framework to programs that rely on both PM and multi-threading to reason about persistency and concurrency simultaneously.
Zoom link: https://usc.zoom.us/j/4326990557
WebCast Link: https://usc.zoom.us/j/4326990557
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Thesis Proposal - Ehsan Qasemi
Fri, Dec 02, 2022 @ 11:00 AM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
Ph.D. Candidate: Ehsan Qasemi
Title: Multi-Modal Preconditioned Commonsense Inference
Committee:
Muhao Chen, Aiichiro Nakano, Daniel O'Leary, Fred Morstatter, Luis Garcia
Humans can seamlessly reason with circumstantial preconditions of commonsense knowledge. We understand that "a glass is used for drinking water", unless "the glass is broken" or "the water is toxic". Despite state-of-the-art (SOTA) models' impressive performance in inferring commonsense knowledge, it is unclear whether they understand the circumstantial preconditions.
In this dissertation, I initially propose a novel challenge of reasoning with preconditions attributed to commonsense knowledge, design three tasks based on the challenge in text-only setup, and show there is a significant gap between SOTA language models' performance and human's on our tasks. I then use weak supervision in a combination of targeted fine-tuning strategies to improve the language model's performance on the preconditioned inference task. Finally, I go beyond the text-only setup and investigate the problem of preconditioned inference in a multi-modal setup when the model is challenged to infer the preconditions from an image.
Zoom link: https://usc.zoom.us/j/92119832136 Date: Friday Dec 2nd, 11AM-12 PM
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense- Sebastien Arnold
Mon, Dec 05, 2022 @ 10:00 AM - 11:30 AM
Thomas Lord Department of Computer Science
University Calendar
Committee: Maja Mataric, Fei Sha, Salman Avestimehr, Jesse Thomason, Stefanos Nikolaidis.
Title: Quickly solving new tasks, with meta-learning and without
- Date: Monday 12/5 at 10am PT on
Zoom: https://usc.zoom.us/j/96456583921?pwd=VHZTbTRZYnAzSkErY2RzenpSY1ZGZz09
- Abstract (shortened):
The success of modern machine learning (ML) stems from the unreasonable effectiveness of large data. But what about niche tasks with limited data? Some methods are able to quickly solve those tasks by first pretraining ML models on many generic tasks in a way that lets them quickly adapt to unseen new tasks. Those methods are known to ``learn how to learn'' and thus fall under the umbrella of meta-learning. While meta-learning can be successful, the inductive biases that enable fast adaptation remain poorly understood.
This thesis takes a first step towards an understanding of meta-learning, and reveals a set of guidelines which help design novel and improved methods for fast adaptation. Our core contribution is a study of the solutions found by meta-learning. We uncover the working principles that let them adapt so quickly: their parameters partition into three groups, one to compute task-agnostic features, another for task-specific features, and a third that accelerates adaptation to new tasks.
Building on those insights we introduce several methods to drastically speed up adaptation.
We propose Kronecker-factored meta-optimizers which significantly improve post-adaptation performance of models that are otherwise too small to meta-learn. We also show how to apply our insights to a visual reinforcement learning setting where meta-learning is impractical. Freezing task-agnostic parameters and adapting task-specific ones with policy-induced self-supervision enables adaptation to unseen tasks with large feature extractors pretrained on generic vision datasets.
WebCast Link: https://usc.zoom.us/j/96456583921?pwd=VHZTbTRZYnAzSkErY2RzenpSY1ZGZz09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Yuchen Lin
Mon, Dec 05, 2022 @ 10:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Yuchen Lin
Title: Evaluating and Improving the Commonsense Reasoning Ability of Language Models
Committee: Xiang Ren (chair), Ram Nevatia, Yan Liu, Toby Mintz
Date & Time: Dec 5th (Monday) from 10:00 AM to 12:00 PM.
Zoom: https://usc.zoom.us/j/91622202680?pwd=cUFNbzY2OXYyTFpuaFVIZHlTWEtLUT09
Abstract:
Large pre-trained language models have become the foundation models for natural language processing. Some LMs (e.g., GPT-3) show the potential to acquire general language intelligence. However, we find that they can still make mistakes because they lack commonsense knowledge and reasoning ability, which are of vital significance in developing human-level general AI systems. In this talk, I will introduce how we can better evaluate and improve the commonsense reasoning (CSR) ability of LMs. Prior works mainly use mask-based probing and multiple-choice QA for evaluation. Their limitations prevent us from comprehensively measuring the CSR ability of LMs. To this end, I will present several benchmarks that aim to measure CSR ability in terms of open-endedness, generalization, and robustness, which are three key dimensions that are missing from the prior evaluation protocols. Then, I will introduce CSR methods that improve LMs by incorporating external knowledge. The external knowledge can be either structured graphs (e.g., ConceptNet) or unstructured text (e.g., GenericsKB), or even implicit as input-output pairs. Finally, I will briefly introduce a few interesting future directions for CSR.
WebCast Link: https://usc.zoom.us/j/91622202680?pwd=cUFNbzY2OXYyTFpuaFVIZHlTWEtLUT09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Thesis Proposal - Jingyao Ren
Mon, Dec 05, 2022 @ 03:30 PM - 04:30 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Jingyao Ren
Committee: Sven Koenig, Gaurav Sukhatme, Stefanos Nikolaidis, Feifei Qian(ECE Department), Nora Ayanian(Brown University)
Title: Algorithm Selection and Empirical Hardness of Multi Agent Pathfinding Problems
Abstract:
Solving the Multi-Agent Path Finding (MAPF) problem optimally is known to be NP-Hard for both make-span and total arrival time minimization. While many algorithms have been developed to solve MAPF problems, there is no dominating optimal MAPF algorithm that works well in all types of problems and no standard guidelines for when to use which algorithm.
In this work, we develop the deep convolutional network MAPFAST (Multi-Agent Path Finding Algorithm SelecTor), which takes a MAPF problem instance and attempts to select the fastest algorithm to use from a portfolio of algorithms. We improve the performance of our model by including single-agent shortest paths in the instance embedding given to our model and by utilizing supplemental loss functions in addition to a classification loss. We evaluate our model on a large and diverse dataset of MAPF instances, showing that it outperforms all individual algorithms in its portfolio as well as the state-of-the-art optimal MAPF algorithm selector. We also provide an analysis of algorithm behavior in our dataset to gain a deeper understanding of optimal MAPF algorithms' strengths and weaknesses to help other researchers leverage different heuristics in algorithm designs. Several ongoing projects are also proposed affiliated with detailed analysis such as using more advanced MAPF instance encoding techniques, Graph Neural Network based approach and utilizing the empirical hardness of MAPF to boost the performance of algorithm selectors.
This proposal will be hosted virtually. Zoom link: https://usc.zoom.us/j/6164522905
WebCast Link: https://usc.zoom.us/j/6164522905
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Thesis Proposal - Wenzuan Zhou
Tue, Dec 13, 2022 @ 09:00 AM - 11:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Wenxuan Zhou
Title: Relation Extraction: Models, Robustness, and Generalization
Chair: Muhao Chen
Committee members: Laurent Itti, Jonathan May, Tianshu Sun, Robin Jia
Abstract: With large amounts of digital text generated every day, it is important to extract structured knowledge automatically from the text. Relation extraction (RE), as one essential step of the solution, aims at identifying relationships among entities in a given piece of text. In this thesis proposal, I will present my work during my Ph.D. on RE from three perspectives: (1) designing effective RE models based on pretrained language models; (2) Improving the robustness of RE models, especially against entity bias; and (3) building data-efficient RE models in low-resource scenarios, which is important for real-world applications. After these, I will introduce my ongoing work and future directions for RE.
WebCast Link: : https://usc.zoom.us/j/6915039300
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense -Sara Mohammadinejad
Wed, Dec 14, 2022 @ 12:00 PM - 02:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Sara Mohammadinejad
Title: Learning logical abstractions from sequential data
Committee: Jyotirmy Deshmukh, Chao Wang, Jesse Thomason, Mukund Raghothaman, Paul Bogdan
Sequential data refers to data where order between successive data-points is important. Time-series data and spatio-temporal data are examples of sequential data. Cyber-physical system applications such as autonomous vehicles, wearable devices, and avionic systems generate a large volume of time-series data. The Internet-of-Things, complex sensor networks, multi-agent cyber-physical systems are all examples of spatially distributed systems that continuously evolve in time, and such systems generate huge amounts of spatio-temporal data. Designers often look for tools to extract high-level information from such data. Traditional machine learning (ML) techniques for sequential data offer several solutions to solve these problems; however, the artifacts trained by these algorithms often lack interpretability. A definition of interpretability by Biran and Cotton is: Models are interpretable if their decisions can be understood by humans.
Formal parametric logic, such as Signal Temporal Logic (STL) and Spatio-temporal Reach and Escape Logic (STREL) are seeing increasing adoption in the formal methods and industrial communities as go-to specification languages for sequential data. Formal parametric logic are machine-checkable, and human-understandable abstractions for sequential data, and they can be used to tackle a variety of learning problems that include but are not limited to classification, clustering and active learning. The use of formal parametric logic in the context of machine learning tasks has seen considerable amount of interest in recent years. We make several significant contributions to this growing body of literature. This dissertation makes five key contributions towards learning formal parametric logic from sequential data. (1) We develop a new technique for learning STL-based classifiers from time-series data and provide a way to systematically explore the space of all STL formulas. (2) We conduct a user study to investigate whether STL formulas are indeed understandable to humans. (3) As an application of our STL-based learning framework, we investigate the problem of mining environment assumptions for cyber-physical system models. (4) We develop the first set of algorithms for logical unsupervised learning of spatio-temporal data and show that our method generates STREL formulas of bounded description complexity. (5) We design an explainable and interactive learning approach to learn from natural language and demonstrations using STL. Finally, we showcase the effectiveness of our approaches on case studies that include but are not limited to urban transportation, automotive, robotics and air quality monitoring.
WebCast Link: https://usc.zoom.us/j/94145108434?pwd=R3M0Smh0ZVp6UkR4S2hiamdhdjlMUT09
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Hannes Leipold
Mon, Dec 19, 2022 @ 09:00 AM - 11:00 PM
Thomas Lord Department of Computer Science
University Calendar
Committee: Greg ver Steeg, Federico M. Spedalieri, Aiichiro Nakano, and Todd A. Brun
Title: Imposing Classical Symmetries on Quantum Operators with Applications to Optimization
Abstract (shortened):
Applying quantum computers to solve combinatorial optimization tasks is one of the most exciting ways to leverage quantum systems for practical computational advantage. Two of the primary paradigms to leverage NISQ quantum computers for such tasks are Quantum Annealing (QA) and the Quantum Alternating Operator Ansatz (QAOA). A typical approach for both would be to map a combinatorial optimization problem with feasibility constraints to an unconstrained quadratic optimization problem. However, it is possible to impose these constraints on the evolution of the quantum system by selecting the quantum operators applied to the system such that the corresponding observables remain invariant under the evolution of the quantum state. We consider such approaches to be imposing classical symmetries on the quantum operators. As such, the task of finding such Hamiltonians or unitaries for a collection of constraints is an important task for tailoring quantum algorithms to optimization problems with feasibility constraints.
In this thesis, we give an algebraic formulation for imposing an arbitrary collection of constraint symmetries on quantum operators. This allows us to describe a general algorithm to solve the corresponding task for linear constraints in polynomial time for bounded weight operators and classify the complexity of several related computational problems.
We then consider a quantum annealing protocol for the problem of combinational circuit fault diagnostics (CCFD) and analyze features of our approach that make it attractive for quantum annealers built to solve this class of combinatorial optimization problems. Next, we consider several QAOA protocols that are tailored to impose different constraint symmetries underlying this problem and study the trade-offs between the protocols. Our results are consistent with the view that tailoring the ansatz of a protocol to match the underlying symmetry of an optimization problem can be benefical to finding solutions with a lower QAOA depth under several parameter optimization schemes.
WebCast Link: https://usc.zoom.us/j/93517166738
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
