May 2, 2018, 12:00 PM
Gene Kim: Corpus Annotation and Inference with Episodic Logic Type Structure
[Wednesday, May 02, 2018 at 12:00 PM in Wegmans Hall 2506]
A growing interest in tasks involving language understanding by the NLP community has led to the need for effective semantic parsing and inference. Modern NLP systems use semantic representations that do not quite fulfill the nuanced needs for language understanding: adequately modeling language semantics, enabling general inferences, and being accurately recoverable. This document proposes a plan to create a semantic parser of an initial form for a logic that balances these needs.
We present a plan for creating a high-precision semantic parser for an underspecified logical form (ULF) from annotated texts and which fits into further processing for ambiguity resolution. The semantic representation is grounded in Episodic Logic (EL) and ULF fully resolves the type structure with respect to EL while leaving further issues such as scope, word sense, and anaphora in a restricted, but unresolved state. We hypothesize that a divide-and-conquer approach to semantic parsing will lead to higher quality semantic analyses by simplifying the problem, both from the perspective of researchers careful to handle linguistic phenomena appropriately and from the perspective of building an accurate semantic parser. In addition to creating this parser, this project aims to quantify the use of parsed ULFs for making inferences directly. ULFs enable structural inferences, including Natural Logic-like inferences, without further resolution, which are useful for downstream tasks such as dialogue and question answering.
May 3, 2018, 03:00 PM
Jacob William Brock: Pareto Optimization for Cache and Memory Management
[Thursday, May 03, 2018 at 3:00 PM in Gavett 206]
Advancements in computing have necessitated commensurate advancements in locality theory and replacement policies. Specifically, there is a need for replacement policies and shared cache management which facilitate minimum performance guarantees for each user, and policies that account for multiple costs associated with the presence or absence of data in a cache. Existing locality theory provides the framework to create such replacement policies with predictable behavior. This dissertation addresses the growing use of shared CPU caches, code-caching of just-in-time (JIT) compiled code on burgeoning mobile devices, and the budding use of multiple main-memory technologies in a single system (heterogeneous memory).
The thesis of this dissertation is that the increasing complexity and diversity of cache and memory management problems can be effectively addressed by extending locality theory to enable Pareto optimization. Pareto optimality indicates that in the presence of multiple objectives (e.g., space and time), no one objective can be improved without degrading another. When the only objective is time (with misses as a proxy), the optimal solution is also Pareto optimal. When both time and space are objectives, each Pareto optimal solution minimizes space, among all possible solutions, at each time point; and it minimizes time at each space point.
The dissertation extends locality theory, and formulates Pareto optimal solutions for shared last-level caches, heterogeneous memories, and caching for JIT compilation. as well as practicable analogs that can be implemented without the need for oracular knowledge. In shared last-level caches, it presents new theory for predicting shared cache demand, a technique for minimizing group misses, and for balancing that objective with individual performance constraints. For mobile JIT compilation, it presents an optimal and a practicable compilation policy to maximize memory efficiency. Finally, in heterogeneous memories, it proposes adaptations of the optimal VMIN replacement policy and its practicable analog, the Working Set replacement policy (called H-VMIN and H-WS, respectively).
Reception to follow at 5:30pm in Wegmans Hall 3201
May 10, 2018, 11:00 AM
Xiaochang Peng: Mapping Natural Language Sentences to Semantic Graphs
[Thursday, May 10, 2018 at 11:00 AM in Wegmans Hall 2506]
In recent years, there has been growing interest in graph representations of semantics as deeper understanding of natural language are increasingly important for user applications such as information extraction, question answering and dialogue systems. Previously, algorithms for producing tree-structured syntax representations given natural language text have been studied extensively, while algorithms for producing semantic graphs are less investigated. In this thesis, we focus on exploring new approaches for mapping natural language sentences to semantic graph structures that can be easily generalized to different semantic graph parsing tasks.
More specifically, we present different modeling frameworks that take as input a sentence, and produce a semantic graph representation encoding meaning of the sentence as the output. First, we present a neural sequence-to-sequence model for semantic graph parsing, where vertices and edge labels in the target graph structure are predicted directly. We further propose two mathematically sound formalisms for modeling the derivation sequences for constructing semantic graph structures. In the first approach, we use a graph grammar called Synchronous Hyperedge Replacement Grammar (SHRG) to construct the sentence, graph pairs. We also propose a general transition framework called cache transition system for parsing to graph structures, where a cache with fixed size m is included to analyze different semantic graphs. We evaluate these approaches on a semantic graph representation called Abstract Meaning Representation (AMR), while these approaches are general to extend to other semantic graph representations.
Reception to follow at 5:00pm in Wegmans Hall 3201