Affirmative safety: An approach to risk management for high-risk AI
Akash Wasil
Joshua Clymer
Emily Dardaman
Simeon Campos
Evan Murphy
Prominent AI experts have suggested that companies developing high-risk AI systems should be required to show that such systems are safe bef… (voir plus)ore they can be developed or deployed. The goal of this paper is to expand on this idea and explore its implications for risk management. We argue that entities developing or deploying high-risk AI systems should be required to present evidence of affirmative safety: a proactive case that their activities keep risks below acceptable thresholds. We begin the paper by highlighting global security risks from AI that have been acknowledged by AI experts and world governments. Next, we briefly describe principles of risk management from other high-risk fields (e.g., nuclear safety). Then, we propose a risk management approach for advanced AI in which model developers must provide evidence that their activities keep certain risks below regulator-set thresholds. As a first step toward understanding what affirmative safety cases should include, we illustrate how certain kinds of technical evidence and operational evidence can support an affirmative safety case. In the technical section, we discuss behavioral evidence (evidence about model outputs), cognitive evidence (evidence about model internals), and developmental evidence (evidence about the training process). In the operational section, we offer examples of organizational practices that could contribute to affirmative safety cases: information security practices, safety culture, and emergency response capacity. Finally, we briefly compare our approach to the NIST AI Risk Management Framework. Overall, we hope our work contributes to ongoing discussions about national and global security risks posed by AI and regulatory approaches to address these risks.
CryCeleb: A Speaker Verification Dataset Based on Infant Cry Sounds
David Budaghyan
Arsenii Gorin
Charles Onu
This paper describes the Ubenwa CryCeleb dataset - a labeled collection of infant cries - and the accompanying CryCeleb 2023 task, which is … (voir plus)a public speaker verification challenge based on cry sounds. We released more than 6 hours of manually segmented cry sounds from 786 newborns for academic use, aiming to encourage research in infant cry analysis. The inaugural public competition attracted 59 participants, 11 of whom improved the baseline performance. The top-performing system achieved a significant improvement scoring 25.8% equal error rate, which is still far from the performance of state-of-the-art adult speaker verification systems. Therefore, we believe there is room for further research on this dataset, potentially extending beyond the verification task.
Directed Scattering for Knowledge Graph-Based Cellular Signaling Analysis
Aarthi Venkat
Joyce Chew
Ferran Cardoso Rodriguez
Christopher J. Tape
Michael Perlmutter
Directed graphs are a natural model for many phenomena, in particular scientific knowledge graphs such as molecular interaction or chemical … (voir plus)reaction networks that define cellular signaling relationships. In these situations, source nodes typically have distinct biophysical properties from sinks. Due to their ordered and unidirectional relationships, many such networks also have hierarchical and multiscale structure. However, the majority of methods performing node- and edge-level tasks in machine learning do not take these properties into account, and thus have not been leveraged effectively for scientific tasks such as cellular signaling network inference. We propose a new framework called Directed Scattering Autoencoder (DSAE) which uses a directed version of a geometric scattering transform, combined with the non-linear dimensionality reduction properties of an autoencoder and the geometric properties of the hyperbolic space to learn latent hierarchies. We show this method outperforms numerous others on tasks such as embedding directed graphs and learning cellular signaling networks.
Focal Modulation Networks for Interpretable Sound Classification
Luca Della Libera
The increasing success of deep neural networks has raised concerns about their inherent black-box nature, posing challenges related to inter… (voir plus)pretability and trust. While there has been extensive exploration of interpretation techniques in vision and language, interpretability in the audio domain has received limited attention, primarily focusing on post-hoc explanations. This paper addresses the problem of interpretability by-design in the audio domain by utilizing the recently proposed attention-free focal modulation networks (FocalNets). We apply FocalNets to the task of environmental sound classification for the first time and evaluate their interpretability properties on the popular ESC-50 dataset. Our method outperforms a similarly sized vision transformer both in terms of accuracy and interpretability. Furthermore, it is competitive against PIQ, a method specifically designed for post-hoc interpretation in the audio domain.
PathOCl: Path-Based Prompt Augmentation for OCL Generation with GPT-4
Seif Abukhalaf
Mohammad Hamdaqa
The rapid progress of AI-powered programming assistants, such as GitHub Copilot, has facilitated the development of software applications. T… (voir plus)hese assistants rely on large language models (LLMs), which are foundation models (FMs) that support a wide range of tasks related to understanding and generating language. LLMs have demonstrated their ability to express UML model specifications using formal languages like the Object Constraint Language (OCL). However, the context size of the prompt is limited by the number of tokens an LLM can process. This limitation becomes significant as the size of UML class models increases. In this study, we intro-duce PathOCL, a novel path-based prompt augmentation technique designed to facilitate OCL generation. PathOCL addresses the limi-tations of LLMs, specifically their token processing limit and the challenges posed by large UML class models. PathOCL is based on the concept of chunking, which selectively augments the prompts with a subset of UML classes relevant to the English specification. Our findings demonstrate that PathOCL, compared to augmenting the complete UML class model (UML-Augmentation), generates a higher number of valid and correct OCL constraints using the GPT-4 model. Moreover, the average prompt size crafted using PathOCL significantly decreases when scaling the size of the UML class models.
Resource-Efficient Separation Transformer
Luca Della Libera
Samuele Cornell
Frédéric Lepoutre
François Grondin
Transformers have recently achieved state-of-the-art performance in speech separation. These models, however, are computationally demanding … (voir plus)and require a lot of learnable parameters. This paper explores Transformer-based speech separation with a reduced computational cost. Our main contribution is the development of the Resource-Efficient Separation Transformer (RE-SepFormer), a self-attention-based architecture that reduces the computational burden in two ways. First, it uses non-overlapping blocks in the latent space. Second, it operates on compact latent summaries calculated from each chunk. The RE-SepFormer reaches a competitive performance on the popular WSJ0-2Mix and WHAM! datasets in both causal and non-causal settings. Remarkably, it scales significantly better than the previous Transformer-based architectures in terms of memory and inference time, making it more suitable for processing long mixtures.
SKILL: Similarity-aware Knowledge distILLation for Speech Self-Supervised Learning
Luca Zampierin
Ghouthi Boukli Hacene
Bac Nguyen
Towards Practical Tool Usage for Continually Learning LLMs
Jerry Huang
Prasanna Parthasarathi
Mehdi Rezagholizadeh
Large language models (LLMs) show an innate skill for solving language based tasks. But insights have suggested an inability to adjust for i… (voir plus)nformation or task-solving skills becoming outdated, as their knowledge, stored directly within their parameters, remains static in time. Tool use helps by offloading work to systems that the LLM can access through an interface, but LLMs that use them still must adapt to nonstationary environments for prolonged use, as new tools can emerge and existing tools can change. Nevertheless, tools require less specialized knowledge, therefore we hypothesize they are better suited for continual learning (CL) as they rely less on parametric memory for solving tasks and instead focus on learning when to apply pre-defined tools. To verify this, we develop a synthetic benchmark and follow this by aggregating existing NLP tasks to form a more realistic testing scenario. While we demonstrate scaling model size is not a solution, regardless of tool usage, continual learning techniques can enable tool LLMs to both adapt faster while forgetting less, highlighting their potential as continual learners.
Why People Contribute Software Documentation
Deeksha M. Arya
Martin P. Robillard
Software technologies are used by a large population of programmers with diverse backgrounds. To fulfill their need for information, enthusi… (voir plus)asts contribute numerous learning resources that vary in style and content, and act as documentation for the corresponding technology. We interviewed 26 volunteer contributors to understand why they create such documentation. We surface five motivations our informants had for contributing documentation, including to overcome issues they had faced with documentation and to capture their own learning. Among other findings, our observations suggest that the unique experience and background of documentation contributors provides the opportunity to create documentation that caters to users who have information needs and preferences similar to that of the contributor. CCS CONCEPTS • Software and its engineering
Why People Contribute Software Documentation
Deeksha M. Arya
Martin P. Robillard
Assessing Numerical Analysis Performance with the Practi Mobile App
Kristin Garn
Raymond J. Spiteri
Model-independent Approach of the JUNO 8B Solar Neutrino Program
Jun Zhao
B. Yue
Haoqi Lu
Yufeng Li
J. Ling
Zeyuan Yu
Angel Abusleme
Thomas Adam
Shakeel Ahmad
Rizwan Ahmed
Sebastiano Aiello
Muhammad Akram
Abid Aleem
Tsagkarakis Alexandros
F. An
Q. An
Giuseppe Andronico
Nikolay Anfimov
Vito Antonelli
Tatiana Antoshkina … (voir 480 de plus)
Burin Asavapibhop
J. Andr'e
Didier Auguste
Weidong Bai
Nikita Balashov
Wander Baldini
Andrea Barresi
Davide Basilico
Eric Baussan
Marco Bellato
Antonio Bergnoli
Thilo Birkenfeld
Sylvie Blin
D. Blum
Simon Blyth
Anastasia Bolshakova
Mathieu Bongrand
Clément Bordereau
Dominique Breton
Augusto Brigatti
Riccardo Brugnera
Riccardo Bruno
Antonio Budano
Jose Busto
I. Butorov
Anatael Cabrera
Barbara Caccianiga
Hao Cai
Xiao Cai
Yanke Cai
Z. Cai
Riccardo Callegari
Antonio Cammi
Agustin Campeny
C. Cao
Guofu Cao
Jun Cao
Rossella Caruso
C. Cerna
Chi Chan
Jinfan Chang
Yun Chang
Guoming Chen
Pingping Chen
Po-An Chen
Shaomin Chen
Xurong Chen
Yixue Chen
Yu Chen
Zhiyuan Chen
Zikang Chen
Jie Cheng
Yaping Cheng
Alexander Chepurnov
Alexey Chetverikov
Davide Chiesa
Pietro Chimenti
Artem Chukanov
Gérard Claverie
Catia Clementi
Barbara Clerbaux
Marta Colomer Molla
Selma Conforti Di Lorenzo
Daniele Corti
Flavio Dal Corso
Olivia Dalager
C. Taille
Z. Y. Deng
Ziyan Deng
Wilfried Depnering
Marco Diaz
Xuefeng Ding
Yayun Ding
Bayu Dirgantara
Sergey Dmitrievsky
Tadeas Dohnal
Dmitry Dolzhikov
Georgy Donchenko
Jianmeng Dong
Evgeny Doroshkevich
Marcos Dracos
Frédéric Druillole
Ran Du
S. X. Du
Stefano Dusini
Martin Dvorak
Timo Enqvist
H. Enzmann
Andrea Fabbri
D. Fan
Lei Fan
Jian Fang
Wen Fang
Marco Fargetta
Dmitry Fedoseev
Zheng-hao Fei
Li-Cheng Feng
Qichun Feng
R. Ford
Amélie Fournier
H. Gan
Feng Gao
Alberto Garfagnini
Arsenii Gavrikov
Marco Giammarchi
Nunzio Giudice
Maxim Gonchar
G. Gong
Hui Gong
Yuri Gornushkin
A. Gottel
Marco Grassi
Maxim Gromov
Vasily Gromov
M. H. Gu
X. Gu
Yunting Gu
M. Guan
Yuduo Guan
Nunzio Guardone
Cong Guo
Jingyuan Guo
Wanlei Guo
Xinheng Guo
Yuhang Guo
Paul Hackspacher
Caren Hagner
Ran Han
Yang Han
Miao He
W. He
Tobias Heinz
Patrick Hellmuth
Yue-kun Heng
Rafael Herrera
Y. Hor
Shaojing Hou
Yee Hsiung
Bei-Zhen Hu
Hang Hu
Jianrun Hu
Jun Hu
Shouyang Hu
Tao Hu
Yuxiang Hu
Zhuojun Hu
Guihong Huang
Hanxiong Huang
Kaixuan Huang
Wenhao Huang
Xinglong Huang
X. T. Huang
Yongbo Huang
Jiaqi Hui
L. Huo
Wenju Huo
Cédric Huss
Safeer Hussain
Ara Ioannisian
Roberto Isocrate
Beatrice Jelmini
Ignacio Jeria
Xiaolu Ji
Huihui Jia
Junji Jia
Siyu Jian
Di Jiang
Wei Jiang
Xiaoshan Jiang
X. Jing
Cécile Jollet
L. Kalousis
Philipp Kampmann
Li Kang
Rebin Karaparambil
Narine Kazarian
Amina Khatun
Khanchai Khosonthongkee
Denis Korablev
K. Kouzakov
Alexey Krasnoperov
Nikolay Kutovskiy
Pasi Kuusiniemi
Tobias Lachenmaier
Cecilia Landini
Sébastien Leblanc
Victor Lebrin
F. Lefèvre
R. Lei
Rupert Leitner
Jason Leung
Daozheng Li
Demin Li
Fei Li
Fule Li
Gaosong Li
Huiling Li
Mengzhao Li
Min Li
Nan Li
Qingjiang Li
Ruhui Li
Ruiting Lei
Shanfeng Li
Tao Li
Teng Li
Weidong Li
Wei-guo Li
Xiaomei Li
Xiaonan Li
Xinglong Li
Yi Li
Yichen Li
Zepeng Li
Zhaohan Li
Zhibing Li
Ziyuan Li
Zonghui Li
Hao Liang
Jiaming Yan
Ayut Limphirat
G. Lin
Shengxin Lin
Tao Lin
Ivano Lippi
Yang Liu
Haidong Liu
Hao Liu
Hongbang Liu
Hongjuan Liu
Hongtao Liu
Hui Liu
Jianglai Liu
Jinchang Liu
Min Liu
Qian Liu
Q. Liu
Runxuan Liu
Shubin Liu
Shulin Liu
Xiaowei Liu
Xiwen Liu
Yan Liu
Yunzhe Liu
Alexey Lokhov
Paolo Lombardi
Claudio Lombardo
K. Loo
Chuan Lu
Jingbin Lu
Junguang Lu
Shuxiang Lu
Bayarto Lubsandorzhiev
Sultim Lubsandorzhiev
Livia Ludhova
Arslan Lukanov
Daibin Luo
F. Luo
Guang Luo
Shu Luo
Wu Luo
Xiaojie Luo
Vladimir Lyashuk
B. Ma
Bing Ma
R. Q. Ma
Si Ma
Xiaoyan Ma
Xubo Ma
Jihane Maalmi
Jingyu Mai
Yury Malyshkin
Roberto Carlos Mandujano
Fabio Mantovani
Francesco Manzali
Xin Mao
Yajun Mao
S. Mari
F. Marini
Cristina Martellini
Gisèle Martin-chassard
Agnese Martini
Matthias Mayer
Davit Mayilyan
Ints Mednieks
Yu Meng
Anselmo Meregaglia
Emanuela Meroni
David J. Meyhofer
Mauro Mezzetto
Jonathan Andrew Miller
Lino Miramonti
Paolo Montini
Michele Montuschi
Axel Muller
M. Nastasi
D. Naumov
Elena Naumova
Diana Navas-Nicolas
Igor Nemchenok
Minh Thuan Nguyen Thi
Alexey Nikolaev
F. Ning
Zhe Ning
Hiroshi Nunokawa
Lothar Oberauer
Juan Pedro Ochoa-Ricoux
Alexander Olshevskiy
Domizia Orestano
Fausto Ortica
Rainer Othegraven
A. Paoloni
Sergio Parmeggiano
Y. P. Pei
Nicomede Pelliccia
Anguo Peng
Yu Peng
Yuefeng Peng
Z-R Peng
Frédéric Perrot
P. Petitjean
Fabrizio Petrucci
Oliver Pilarczyk
Luis Felipe Piñeres Rico
Artyom Popov
Pascal Poussot
Ezio Previtali
Fazhi Qi
M. Qi
Sen Qian
X. Qian
Zhen Qian
Hao-xue Qiao
Zhonghua Qin
S. Qiu
Gioacchino Ranucci
Neill Raper
A. Re
Henning Rebber
Abdel Rebii
Mariia Redchuk
Bin Ren
Jie Ren
Barbara Ricci
Mariam Rifai
Mathieu Roche
Narongkiat Rodphai
Aldo M. Romani
Bedřich Roskovec
X. Ruan
Arseniy Rybnikov
Andrey Sadovsky
Paolo Saggese
Simone Sanfilippo
Anut Sangka
Utane Sawangwit
Julia Sawatzki
Michaela Schever
Cédric Schwab
Konstantin Schweizer
Alexandr Selyunin
Andrea Serafini
Giulio Settanta
M. Settimo
Zhuang Shao
V. Sharov
Arina Shaydurova
Jingyan Shi
Yanan Shi
Vitaly Shutov
Andrey Sidorenkov
Fedor Šimkovic
Chiara Sirignano
Jaruchit Siripak
Monica Sisti
Maciej Slupecki
Mikhail Smirnov
Oleg Smirnov
Thiago Sogo-Bezerra
Sergey Sokolov
Julanan Songwadhana
Boonrucksar Soonthornthum
Albert Sotnikov
Ondvrej vSr'amek
Warintorn Sreethawong
A. Stahl
Luca Stanco
Konstantin Stankevich
Duvsan Vstef'anik
Hans Steiger
Jochen Steinmann
Tobias Sterr
M. Stock
Virginia Strati
Alexander Studenikin
Jun Su
Shifeng Sun
Xilei Sun
Yongjie Sun Sun
Yongzhao Sun
Zhengyang Sun
Narumon Suwonjandee
Michal Szelezniak
Qiang Tang
Quan Tang
Xiao Tang
Alexander Tietzsch
Igor Tkachev
Tomas Tmej
M. Torri
K. Treskov
Andrea Triossi
Giancarlo Troni
Wladyslaw Trzaska
Cristina Tuve
Nikita Ushakov
Vadim Vedin
Giuseppe Verde
Maxim Vialkov
Benoit Viaud
Cornelius Moritz Vollbrecht
C. Volpe
Katharina von Sturm
Vit Vorobel
Dmitriy Voronin
Lucia Votano
Pablo Walker
Caishen Wang
Chung-Hsiang Wang
En Wang
Guoli Wang
Jian Wang
Jun Wang
Lucinda W. Wang
Meifen Wang
Meng Wang
Ruiguang Wang
Siguang Wang
Wei Wang
Wenshuai Wang
Xi Wang
Xiangyue Wang
Yangfu Wang
Yaoguang Wang
Yi Xing Wang
Yifang Wang
Yuanqing Wang
Yuman Wang
Zhe Wang
Zheng Wang
Zhimin Wang
Zongyi Wang
Apimook Watcharangkool
Wei Wei
Wenlu Wei
Yadong Wei
K. Wen
Kaile Wen
Christopher Wiebusch
S. Wong
Bjoern Wonsak
Diru Wu
Qun Wu
Zhi Wu
Michael Wurm
Jacques Wurtz
Christian Wysotzki
Yufei Xi
D. Xia
Xiang Xiao
Xiaochuan Xie
Yu-guang Xie
Z. P. Xie
Zhao-Liang Xin
Z. Xing
Benda D. Xu
Chengze Xu
Donglian Xu
Fanrong Xu
The physics potential of detecting 8B solar neutrinos will be exploited at the Jiangmen Underground Neutrino Observatory (JUNO), in a model-… (voir plus)independent manner by using three distinct channels of the charged current (CC), neutral current (NC), and elastic scattering (ES) interactions. Due to the largest-ever mass of 13C nuclei in the liquid scintillator detectors and the expected low background level, 8B solar neutrinos are observable in the CC and NC interactions on 13C for the first time. By virtue of optimized event selections and muon veto strategies, backgrounds from the accidental coincidence, muon-induced isotopes, and external backgrounds can be greatly suppressed. Excellent signal-to-background ratios can be achieved in the CC, NC, and ES channels to guarantee the observation of the 8B solar neutrinos. From the sensitivity studies performed in this work, we show that JUNO, with 10 yr of data, can reach the 1σ precision levels of 5%, 8%, and 20% for the 8B neutrino flux, sin 2 θ 12 , and Δ m 21 2 , respectively. Probing the details of both solar physics and neutrino physics would be unique and helpful. In addition, when combined with the Sudbury Neutrino Observatory measurement, the world's best precision of 3% is expected for the measurement of the 8B neutrino flux.