Select a calendar:
Filter January Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for January
-
PhD Dissertation Defense - Weizhao Jin
Thu, Jan 16, 2025 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Efficiency in Privacy-Preserving Computation via Domain Knowledge
Date and Time: Thursday, January 16th, 2025, from 11:00 AM to 12:00 PM
Location: THH 221
Committee: Srivatsan Ravi (CS), Bhaskar Krishnamachari (ECE), Harsha V. Madhyastha (CS), Fred Morstatter (CS)
Abstract: In recent years, the growing reliance on user data for building server-side applications and services has significantly heightened the importance of data privacy. To meet expanding privacy regulations like GDPR, service providers have turned to privacy-preserving methods that maintain computational functionality while protecting user privacy. However, integrating techniques such as homomorphic encryption into application protocols presents a critical challenge: achieving a balance between privacy and efficiency. This thesis explores two distinct domains within privacy-preserving computation, offering practical, domain-specific solutions to address challenges related to overheads and protocol complexity. The focus is on achieving efficient privacy in both machine learning and networks/IoT. To illustrate how leveraging domain-specific insights—from federated learning, entity resolution, and computer networking—can substantially enhance the efficiency of privacy-preserving computation, we first introduce a selective encryption strategy for large-scale federated learning models, reducing overhead by encrypting only sensitive parameters while still maintaining robust privacy guarantees; secondly, we demonstrate how homomorphic encryption can be optimized for deep entity resolution via a two-stage computation scheme and novel techniques including synthetic ranging and polynomial degree optimization that preserve accuracy under encrypted computation; finally, we apply Non-Interactive Zero-Knowledge proofs to achieve lightweight privacy-preserving path validation across multi-authority network slices, ensuring data forwarding compliance without revealing sensitive topology details by utilizing a backward pairwise validation procedure. Taken together, these studies highlight how targeting domain-specific challenges via domain-specific knowledge can yield practical, scalable frameworks for efficient privacy-preserving computation.
Zoom Link: https://usc.zoom.us/j/99543392059?pwd=FlQxFqagihzPzEV4tfBaemgHBwOwUM.1Location: Mark Taper Hall Of Humanities (THH) - 221
Audiences: Everyone Is Invited
Contact: Weizhao Jin
-
PhD Dissertation Defense - Mehrnoosh Mirtaheri
Wed, Jan 22, 2025 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Toward Learning and Forecasting with Temporal Knowledge Graphs
Date and Time: Tuesday, January 22nd, 2025 - 1:00p - 3:00p
Location: SAL 213
Committee: Aram Galstyan (Chair), Emilio Ferrara (Tenured Faculty), Fred Morstatter, Antonio Ortega (External Faculty
Abstract: Temporal knowledge graphs (TKGs) model real-world relationships between entities over time, enabling insight extraction from unstructured data. While powerful for various applications, TKGs are inherently limited by incompleteness and noise, making their completion and forecasting crucial research areas.
This thesis tackles the key challenges in TKG forecasting: relation sparsity in large-scale graphs, continuous integration of new data while preserving existing knowledge, and entity evolution as new entities emerge and existing ones appear in novel contexts. Through novel methodological frameworks, this research demonstrates improved predictive accuracy, robustness to data sparsity, and adaptability to evolving data, validated through extensive evaluation on both standard benchmark and real-world datasets.
Zoom Link: https://usc.zoom.us/j/96220815599Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Mehrnoosh Mirtaheri
-
PhD Dissertation Defense - Mozhdeh Gheini
Fri, Jan 24, 2025 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
University Calendar
Title: Inductive Biases for Data- and Parameter-Efficient Transfer Learning
Date and Time: Fri, Jan 24, 2025 @ 10:00 AM - 12:00 PM
Location: Salvatori Computer Science Center (SAL) - 213 and https://usc.zoom.us/j/6564802162
Committee Members: Jonathan May (Chair), Emilio Ferrara, Xuezhe Ma, Khalil Iskarous
Abstract: Data- and resource-intensive pre-training and fine-tuning applied upon Transformer-based models is the dominant paradigm at the forefront of rapid advancements in natural language processing, human language technologies, and most notably, large language models. Such reliance on massive amounts of data, computation, and energy, while effective and impressive from a performance-only perspective, can hinder open, nonexclusive, and sustainable development of these technologies. In this talk, we present how certain inductive biases can be devised to adjust current natural language methods under resource-constrained scenarios and provide insights into why the proposed inductive biases are successful in such cases.
Specifically, we discuss four research directions on data and parameter efficiency of fine-tuning and transfer learning in natural language processing: (1) a universal regimen that creates a single pre-trained checkpoint suitable for machine translation transfer to practically any language pair and eliminates the need for ad hoc pre-training; (2) an architecture-guided parameter-efficient fine-tuning method that performs competitively with full fine-tuning while exclusively updating cross-attention parameters; (3) an analysis of MEGA, a recently introduced augmentation of the Transformer architecture to incorporate explicit recency bias, through the lens of transfer learning; and (4) a meta-learning algorithm to prime pre-trained models for specific fine-tuning strategies.
Combined with ablations that show how they are effective and analyses that demonstrate their generalizability, these directions are meant to serve as tools for resource-efficient transfer learning for natural language processing.Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Mozhdeh Gheini
