Beyond fine-tuning: LoRA modules boost near-OOD detection and LLM security

Chat with Graph Search

Ask any question about EPFL courses, lectures, exercises, research, news, etc. or try the example questions below.

DISCLAIMER: The Graph Chatbot is not programmed to provide explicit or categorical answers to your questions. Rather, it transforms your questions into API requests that are distributed across the various IT services officially administered by EPFL. Its purpose is solely to collect and recommend relevant references to content that you can explore to help you answer your questions.

Under resource constraints, LLMs are usually fine-tuned with additional knowledge using Parameter Efficient Fine-Tuning (PEFT), using Low-Rank Adaptation (LoRA) modules. In fact, LoRA injects a new set of small trainable matrices to adapt an LLM to a new t ...

Large language models (LLMs) have shown strong performance in tasks across domains but struggle with chemistry-related problems. These models also lack access to external knowledge sources, limiting their usefulness in scientific applications. We introduce ...

Traditional martial arts are treasures of humanity's knowledge and critical carriers of sociocultural memories throughout history. However, such treasured practices have encountered various challenges in knowledge transmission and now feature many entries ...

The ability to reason, plan and solve highly abstract problems is a hallmark of human intelligence. Recent advancements in artificial intelligence, propelled by deep neural networks, have revolutionized disciplines like computer vision and natural language ...

In this dissertation, we propose multiple methods to improve transfer learning for pretrained language models (PLMs). Broadly, transfer learning is a powerful technique in natural language processing, where a language model is first pre-trained on a data-r ...

Two related methods for inverting line-integrated measurements are presented in this research paper in the context of the recent deuterium-tritium experiments in the JET tokamak. Unlike traditional methods of tomography, these methods rely on making use of ...

Purpose: Despite being an authentic carrier of various cultural practices, the human body is often underutilised to access the knowledge of human body. Digital inventions today have created new avenues to open up cultural data resources, yet mainly as appa ...

Visual perception is indispensable for many real-world applications. However, perception models deployed in the real world will encounter numerous and unpredictable distribution shifts, for example, changes in geographic locations, motion blur, and adverse ...

Various endeavours into semantic web technologies and ontology engineering have been made within the organisation of cultural data, facilitating public access to digital assets. Although models for conceptualising objects have reached a certain level of ma ...

Time has always been a central factor in understanding the challenges of daily mobility. For a long time, and still today, methods of economic evaluation of transport projects have monetized time savings so that they can be included in the cost–benefit ana ...

Linear-Quadratic-Gaussian (LQG) control is a fundamental control paradigm that is studied in various fields such as engineering, computer science, economics, and neuroscience. It involves controlling a system with linear dynamics and imperfect observations ...

Investigating the intangible nature of a cultural domain can take multiple forms, addressing for example the aesthetic, epistemic and social dimensions of its phenomenology. The context of Southern Chinese martial arts is of particular significance as it c ...

Large language models (LLMs) have demonstrated human-level performance on a vast spectrum of natural language tasks. However, it is largely unexplored whether they can better internalize knowledge from a structured data, such as a knowledge graph, or from ...

Current task-oriented dialog (TOD) systems mostly manage structured knowledge (e.g. databases and tables) to guide the goal-oriented conversations. However, they fall short of handling dialogs which also involve unstructured knowledge (e.g. reviews and doc ...

As distribution shifts are inescapable in realistic clinical scenarios due to inconsistencies in imaging protocols, scanner vendors, and across different centers, well-trained deep models incur a domain generalization problem in unseen environments. Despit ...

A Concideration of Typologies for Housing the Sick as a Spatial Manifestation of knowledge. ...

With the application requirements of wireless technology in implantable bioelectronics, knowledge of the fundamental limits for implanted antennas becomes critical. In this work, we investigated the variation of maximum power density within simplified body ...

Test-time domain adaptation aims to adapt a source pretrained model to a target domain without using any source data. Existing works mainly consider the case where the target domain is static. However, real-world machine perception systems are running in n ...

Action-oriented, transformative, and transdisciplinary approaches are increasingly heralded as promising tools that enable researchers not only to produce new knowledge about a situation, but also to actively engage in tackling it; however, such approaches ...

In this thesis, we present a theory of value for the design and analysis of service ecosystems. The theory is based on general systems thinking. The concept of a system is used for relating knowledge from different disciplines (such as software engineering ...

Beyond fine-tuning: LoRA modules boost near-OOD detection and LLM security

Chat with Graph Search

Beyond fine-tuning: LoRA modules boost near-OOD detection and LLM security

Augmenting large language models with chemistry tools

Ontology-based Knowledge Representation for Traditional Martial Arts

Infusing structured knowledge priors in neural models for sample-efficient symbolic reasoning

Improving Generalization of Pretrained Language Models

Regressive Tomography and Direct Regressive Reconstruction: A Research Note

Unlocking a multimodal archive of Southern Chinese martial arts through embodied cues

Making Computer Vision Models Robust and Adaptive

Building a Knowledge Graph of Chinese Kung Fu Masters From Heterogeneous Bilingual Data

Appropriations and Uses of Travel Time

Distributionally Robust Linear Quadratic Control

The Facets of Intangible Heritage in Southern Chinese Martial Arts: Applying a Knowledge-Driven Cultural Contact Detection Approach

SKILL: Structured Knowledge Infusion for Large Language Models

End-to-End Task-Oriented Dialog Modeling With Semi-Structured Knowledge Management

OptTTA: Learnable Test-Time Augmentation for Source-Free Medical Image Segmentation Under Domain Shift

Horizontal Healing

On the Maximum Power Density of Implanted Antennas within Simplified Body Phantoms

Continual Test-Time Domain Adaptation

Explore, engage, empower: methodological insights into a transformative mixed methods study tackling the COVID-19 lockdown

A Theory of Value for Service Ecosystems