A5052833772- Natural Language Processing Techniques
- Topic Modeling
- Speech Recognition and Synthesis
- Multimodal Machine Learning Applications
- Artificial Intelligence in Games
- Speech and dialogue systems
- Linguistic Variation and Morphology
- Authorship Attribution and Profiling
- Reinforcement Learning in Robotics
- Diet and metabolism studies
- Metabolism and Genetic Disorders
- Semantic Web and Ontologies
- Text and Document Classification Technologies
- Renal function and acid-base balance
- Software Engineering Research
- Explainable Artificial Intelligence (XAI)
- Language, Metaphor, and Cognition
- Retinal Imaging and Analysis
- Domain Adaptation and Few-Shot Learning
- Model Reduction and Neural Networks
- Influenza Virus Research Studies
- Imbalanced Data Classification Techniques
- Mental Health Research Topics
- Advanced Text Analysis Techniques
- Music and Audio Processing
Expert System (Italy)
2024
University of Sussex
2023
University of Siena
2016-2022
University of Florence
2018-2021
Florence (Netherlands)
2020-2021
Vision-language models (VLMs), which process image and text inputs, are increasingly integrated into chat assistants other consumer AI applications. Without proper safeguards, however, VLMs may give harmful advice (e.g. how to self-harm) or encourage unsafe behaviours consume drugs). Despite these clear hazards, little work so far has evaluated VLM safety the novel risks created by multimodal inputs. To address this gap, we introduce MSTS, a Multimodal Safety Test Suite for VLMs. MSTS...
The common practice in Machine Learning research is to evaluate the top-performing models based on their performance. However, this often leads overlooking other crucial aspects that should be given careful consideration. In some cases, performance differences between various approaches may insignificant, whereas factors like production costs, energy consumption, and carbon footprint taken into account. Large Language Models (LLMs) are widely used academia industry address NLP problems....
We consider a scenario where an artificial agent is reading stream of text composed set narrations, and it informed about the identity some individuals that are mentioned in portion currently being read. The expected to learn follow thus disambiguating mentions discovering new individuals. focus on case which entities relations propose end-to-end trainable memory network learns discover disambiguate them online manner, performing one-shot learning dealing with small number sparse...
Real-world business applications require a trade-off between language model performance and size. We propose new method for compression that relies on vocabulary transfer. evaluate the various vertical domains downstream tasks. Our results indicate transfer can be effectively used in combination with other techniques, yielding significant reduction size inference time while marginally compromising performance.
Most Machine Learning research evaluates the best solutions in terms of performance. However, race for performing model, many important aspects are often overlooked when, on contrary, they should be carefully considered. In fact, sometimes gaps performance between different approaches neglectable, whereas factors such as production costs, energy consumption, and carbon footprint must take into consideration. Large Language Models (LLMs) extensively adopted to address NLP problems academia...
This report provides an introduction to some Machine Learning tools within the most common development environments. It mainly focuses on practical problems, skipping any theoretical introduction. is oriented both students trying approach and experts looking for new frameworks.
MOTIVATION Alkaptonuria (AKU) is a rare and genetic disease which causes discoloration of bone (a process called ‘ochronosis’) induces early-onset osteoarthritis. AKU data have not been organized yet the has no approved biomarkers. The ability to collect, integrate analyze relevant streams core for developing “Precision Medicine Ecosystem” AKU-dedicated in biological resources are shared between researchers, clinicians patients. Computational modeling can be useful guide generate an...
MOTIVATION Alkaptonuria (AKU) is a rare and genetic disease which causes discoloration of bone (a process called ‘ochronosis’) induces early-onset osteoarthritis. AKU data have not been organized yet the has no approved biomarkers. The ability to collect, integrate analyze relevant streams core for developing “Precision Medicine Ecosystem” AKU-dedicated in biological resources are shared between researchers, clinicians patients. Computational modeling can be useful guide generate an...
In the last decades, deep learning approaches achieved impressive results in many research fields, such as Computer Vision and Natural Language Processing (NLP). NLP particular has greatly benefit from unsupervised methods that allow to learn distributed representation of language. On race for better performances Models have reached hundred billions parameters nowadays. Despite remarkable results, models are still far being fully exploited real world applications. Indeed, these black-boxes,...
Large Language Models have proven highly successful at modelling a variety of tasks. However, this comes steep computational cost that hinders wider industrial uptake. In paper, we present MWT: Multi-Word Tokenizer goes beyond word boundaries by representing frequent multi-word expressions as single tokens. MWTs produce more compact and efficient tokenization yields two benefits: (1) Increase in performance due to greater coverage input data given fixed sequence length budget; (2) Faster...
Crossword puzzles are popular linguistic games often used as tools to engage students in learning. Educational crosswords characterized by less cryptic and more factual clues that distinguish them from traditional crossword puzzles. Despite there exist several publicly available clue-answer pair databases for crosswords, educational pairs datasets missing. In this article, we propose a methodology build clue generation can be instruct Large Language Models (LLMs). By gathering Wikipedia...
Understanding human attention is crucial for vision science and AI. While many models exist free-viewing, less known about task-driven image exploration. To address this, we introduce CapMIT1003, a dataset with captions click-contingent explorations, to study during the captioning task. We also present NevaClip, zero-shot method predicting visual scanpaths by combining CLIP NeVA algorithms. NevaClip generates fixations align representations of foveated stimuli captions. The simulated...
Traditional approaches to Named Entity Recognition (NER) frame the task into a BIO sequence labeling problem. Although these systems often excel in downstream at hand, they require extensive annotated data and struggle generalize out-of-distribution input domains unseen entity types. On contrary, Large Language Models (LLMs) have demonstrated strong zero-shot capabilities. While several works address Zero-Shot NER English, little has been done other languages. In this paper, we define an...
Large language models present opportunities for innovative Question Answering over Knowledge Graphs (KGQA). However, they are not inherently designed query generation. To bridge this gap, solutions have been proposed that rely on fine-tuning or ad-hoc architectures, achieving good results but limited out-of-domain distribution generalization. In study, we introduce a novel approach called Dynamic Few-Shot Learning (DFSL). DFSL integrates the efficiency of in-context learning and semantic...
Recently, several specialized instruction-tuned Large Language Models (LLMs) for Named Entity Recognition (NER) have emerged. Compared to traditional NER approaches, these models strong generalization capabilities. Existing LLMs mainly focus on zero-shot in out-of-domain distributions, being fine-tuned an extensive number of entity classes that often highly or completely overlap with test sets. In this work instead, we propose SLIMER, approach designed tackle never-seen-before named tags by...
Paraphrasing is the task of re-writing an input text using other words, without altering meaning original content. Conversational systems can exploit automatic paraphrasing to make conversation more natural, e.g., talking about a certain topic different paraphrases in time instants. Recently, automatically generating has been approached context Natural Language Generation (NLG). While many existing simply consist rule-based models, recent success Deep Neural Networks several NLG tasks...
Understanding the mechanisms underlying human attention is a fundamental challenge for both vision science and artificial intelligence. While numerous computational models of free-viewing have been proposed, less known about task-driven image exploration. To address this gap, we present CapMIT1003, database captions click-contingent explorations collected during captioning tasks. CapMIT1003 based on same stimuli from well-known MIT1003 benchmark, which eye-tracking data under conditions...
Transformers have been established as one of the most effective neural approach in performing various Natural Language Processing tasks. However, following common trend modern deep architectures, their scale has quickly grown to an extent that reduces concrete possibility for several enterprises train such models from scratch. Indeed, despite high-level performances, general drawback requiring a huge amount training data, computational resources and energy consumption be successfully...
Most Machine Learning research evaluates the best solutions in terms of performance. However, race for performing model, many important aspects are often overlooked when, on contrary, they should be carefully considered. In fact, sometimes gaps performance between different approaches neglectable, whereas factors such as production costs, energy consumption, and carbon footprint must take into consideration. Large Language Models (LLMs) extensively adopted to address NLP problems academia...