A5066490162- Topic Modeling
- Natural Language Processing Techniques
- Gait Recognition and Analysis
- Domain Adaptation and Few-Shot Learning
- Speech and dialogue systems
- Time Series Analysis and Forecasting
- Advanced Image and Video Retrieval Techniques
- Game Theory and Applications
- Wikis in Education and Collaboration
- Image Retrieval and Classification Techniques
- Merger and Competition Analysis
- Neural Networks and Applications
- Indoor and Outdoor Localization Technologies
- Computational Physics and Python Applications
- Advanced Text Analysis Techniques
- Interactive and Immersive Displays
- Business Strategy and Innovation
- Tactile and Sensory Interactions
- Context-Aware Activity Recognition Systems
- Geophysical Methods and Applications
University of Massachusetts Amherst
2020-2023
The vast majority of retrieval models depend on vector inner products to produce a relevance score between query and document. This naturally limits the expressiveness that can be employed. We propose new paradigm, instead producing represent we small neural network which acts as learned function. takes in representation document, this paper use single vector, produces scalar score. To little hypernetwork, weights other networks, our encoder or call it Hypencoder. Experiments in-domain...
Evaluating personalized text generated by large language models (LLMs) is challenging, as only the LLM user, i.e., prompt author, can reliably assess output, but re-engaging same individuals across studies infeasible. This paper addresses challenge of evaluating generation introducing ExPerT, an explainable reference-based evaluation framework. ExPerT leverages to extract atomic aspects and their evidence from reference texts, match aspects, evaluate alignment based on content writing style...
Recent deployment of efficient billion-scale approximate nearest neighbor (ANN) search algorithms on GPUs has motivated information retrieval researchers to develop neural ranking models that learn low-dimensional dense representations for queries and documents use ANN retrieval. However, optimizing these poses several challenges including negative sampling (pair-wise) training. A recent model, called ANCE, successfully uses dynamic using search. This paper improves upon ANCE by proposing a...
In this paper we present MechanoBeat, a 3D printed mechanical tag that oscillates at unique frequency upon user interaction. With the help of an ultra-wideband (UWB) radar array, MechanoBeat can unobtrusively monitor interactions with both stationary and mobile objects. consists small, scalable, easy-to-install tags do not require any batteries, silicon chips, or electronic components. Tags be produced using commodity desktop printers cheap materials. We develop efficient signal processing...
This paper explores SynTOD, a new synthetic data generation approach for developing end-to-end Task-Oriented Dialogue (TOD) Systems capable of handling complex tasks such as intent classification, slot filling, conversational question-answering, and retrieval-augmented response generation, without relying on crowdsourcing or real-world data. SynTOD utilizes state transition graph to define the desired behavior TOD system generates diverse, structured conversations through random walks...
Abstract Duopolies are one of the simplest economic situations where interactions between firms determine market behavior. The standard model a price-setting duopoly is Bertrand model, which has unique solution that both set their prices equal to costs—a paradoxical result obtain zero profit, generally not observed in real duopolies. Here we propose new game theory for duopoly, show resolves behavior and provides consistent general
Knowing how and when people interact with their surroundings is crucial for constructing dynamic intelligent environments. Despite the importance of this problem, an attainable simple solution still lacking. Current solutions often require powered sensors on monitored objects or users themselves. Many such systems use batteries [1-3], which are costly time consuming to replace. Some connect grid, may save swapping batteries, but at price restricted placement options. Other passive tags no...
Many applications require generation of summaries tailored to the user’s information needs, i.e., their intent. Methods that express intent via explicit user queries fall short when query interpretation is subjective. Several datasets exist for summarization with objective intents where, each document and (e.g., “weather”), a single summary suffices all users. No exist, however, subjective “interesting places”) where different users will provide summaries. We present SUBSUME, first dataset...