Can Safety Fine-Tuning Be More Principled? Lessons Learned from Cybersecurity
David Williams-King
Linh Le
Adam Oberman
Yoshua Bengio
As LLMs develop increasingly advanced capabilities, there is an increased need to minimize the harm that could be caused to society by certa… (see more)in model outputs; hence, most LLMs have safety guardrails added, for example via fine-tuning. In this paper, we argue the position that current safety fine-tuning is very similar to a traditional cat-and-mouse game (or arms race) between attackers and defenders in cybersecurity. Model jailbreaks and attacks are patched with bandaids to target the specific attack mechanism, but many similar attack vectors might remain. When defenders are not proactively coming up with principled mechanisms, it becomes very easy for attackers to sidestep any new defenses. We show how current defenses are insufficient to prevent new adversarial jailbreak attacks, reward hacking, and loss of control problems. In order to learn from past mistakes in cybersecurity, we draw analogies with historical examples and develop lessons learned that can be applied to LLM safety. These arguments support the need for new and more principled approaches to designing safe models, which are architected for security from the beginning. We describe several such approaches from the AI literature.
2025-01-19
ArXiv (preprint)
arxiv.org
arxiv.org
Supervised Large Neighbourhood Search for MIPs
Charly Robinson La Rocca
Jean-François Cordeau
Emma Frejinger
Large Neighbourhood Search (LNS) is a powerful heuristic framework for solving Mixed-Integer Programming (MIP) problems. However, designing … (see more)effective variable selection strategies in LNS remains challenging, especially for diverse sets of problems. In this paper, we propose an approach that integrates Machine Learning (ML) within the destroy operator of LNS for MIPs with a focus on minimal offline training. We implement a modular LNS matheuristic as a test bench to compare different LNS heuristics, including our ML-enhanced LNS. Experimental results on the MIPLIB 2017 dataset demonstrate that the matheuristic can significantly improve the performance of state-of-the-art solvers like Gurobi and SCIP. We conduct analyses on noisy oracles to explore the impact of prediction accuracy on solution quality. Additionally, we develop techniques to enhance the ML model through loss adjustments and sampling routines. Our findings suggest that while random LNS remains competitive, our Supervised LNS (SLNS) outperforms other baselines and helps set the foundation for future research on ML for LNS methods that are both efficient and general.
2025-01-18
ArXiv (preprint)
arxiv.org
arxiv.org
Supervised Large Neighbourhood Search for MIPs
Charly Robinson La Rocca
Jean-François Cordeau
Emma Frejinger
Large Neighbourhood Search (LNS) is a powerful heuristic framework for solving Mixed-Integer Programming (MIP) problems. However, designing … (see more)effective variable selection strategies in LNS remains challenging, especially for diverse sets of problems. In this paper, we propose an approach that integrates Machine Learning (ML) within the destroy operator of LNS for MIPs with a focus on minimal offline training. We implement a modular LNS matheuristic as a test bench to compare different LNS heuristics, including our ML-enhanced LNS. Experimental results on the MIPLIB 2017 dataset demonstrate that the matheuristic can significantly improve the performance of state-of-the-art solvers like Gurobi and SCIP. We conduct analyses on noisy oracles to explore the impact of prediction accuracy on solution quality. Additionally, we develop techniques to enhance the ML model through loss adjustments and sampling routines. Our findings suggest that while random LNS remains competitive, our Supervised LNS (SLNS) outperforms other baselines and helps set the foundation for future research on ML for LNS methods that are both efficient and general.
2025-01-18
ArXiv (preprint)
arxiv.org
arxiv.org
Multi-center benchmarking of cervical spinal cord RF coils for 7 T MRI: A traveling spines study
Eva Alonso‐Ortiz
Daniel Papp
Robert L. Barry
Kyota Poëti
Alan C. Seifert
Kyle M. Gilbert
Nibardo Lopez‐Rios
Jan Paska
Falk Eippert
N. Weiskopf
Laura Beghini
Nadine Graedel
Robert Trampel
M. F. Callaghan
Christoph S Aigner
Patrick Freund
Maryam Seif
A. Destruel
Virginie Callot
Johanna Vannesjo … (see 1 more)
Julien Cohen-Adad
Purpose The depth within the body, small diameter, long length, and varying tissue surrounding the spinal cord impose specific consideration… (see more)s when designing radiofrequency coils. The optimal coil configuration for 7 T cervical spinal cord MRI is unknown and, currently, there are very few coil options. The purpose of this work was (1) to establish a quality control protocol for evaluating 7 T cervical spinal cord coils and (2) to use that protocol to evaluate the performance of 4 different coil designs. Methods Three healthy volunteers and a custom anthropomorphic phantom (the traveling spines cohort) were scanned at seven 7 T imaging centers using a common protocol and each center’s specific cervical spinal cord coil. Four different coil designs were tested (two in-house, one Rapid Biomedical, and one MRI.TOOLS design). Results The Rapid Biomedical coil was found to have the highest B1+ efficiency, whereas one of the in-house designs (NeuroPoly Lab) had the highest SNR and the largest spinal cord coverage. The MRI.TOOLS coil had the most uniform B1+ profile along the cervical spinal cord; however, it was limited in its ability to provide the requested flip angles (especially for larger individuals). The latter was also the case for the second in-house coil (MSSM). Conclusion The results of this study serve as a guide for the spinal cord MRI community in selecting the most suitable coil based on specific requirements and offer a standardized protocol for assessing future coils.
2025-01-17
bioRxiv (preprint)
doi.org
Multi-center benchmarking of cervical spinal cord RF coils for 7 T MRI: A traveling spines study
Eva Alonso‐Ortiz
Daniel Papp
Robert L. Barry
Kyota Poëti
Alan C. Seifert
Kyle M. Gilbert
Nibardo Lopez‐Rios
Jan Paska
Falk Eippert
Nikolaus Weiskopf
Laura Beghini
Nadine Graedel
Robert Trampel
Martina F Callaghan
Christoph S. Aigner
Patrick Freund
Maryam Seif
Aurélien Destruel
Virginie Callot
Johanna Vannesjo … (see 1 more)
Julien Cohen-Adad
2025-01-17
bioRxiv (preprint)
doi.org
Maximizing Data and Hardware Reuse for HLS with Early-Stage Symbolic Partitioning
Tzung-Han Juang
Christophe Dubach
While traditional HLS (High-Level Synthesis) converts “high-level” C-like programs into hardware automatically, producing high-performan… (see more)ce designs still requires hardware expertise. Optimizations such as data partitioning can have a large impact on performance since they directly affect data reuse patterns and the ability to reuse hardware. However, optimizing partitioning is a difficult process since minor changes in the parameter choices can lead to totally unpredictable performance. Functional array-based languages have been proposed instead of C-based approaches, as they offer stronger performance guarantees. This paper proposes to follow a similar approach and exposes a divide-and-conquer primitive at the algorithmic level to let users partition any arbitrary computation. The compiler is then free to explore different partition shapes to maximize both data and hardware reuse automatically. The main challenge remains that the impact of partitioning is only known much later in the compilation flow. This is due to the hard-to-predict effects of the many optimizations applied during compilation. To solve this problem, the partitioning is expressed using a set of symbolic tunable parameters, introduced early in the compilation pipeline. A symbolic performance model is then used in the last compilation stage to predict performance based on the possible values of the tunable parameters. Using this approach, a design space exploration is conducted on an Intel Arria 10 FPGAs (Field Programmable Gate Arrays), and competitive performance is achieved on the classical VGG and TinyYolo neural networks.
2025-01-16
ACM Transactions on Architecture and Code Optimization (TACO) (published)
doi.org
Robin: a Suite of Multi-Scale Vision-Language Models and the CHIRP Evaluation Benchmark
Alexis Roger
Prateek Humane
Daniel Z Kaplan
Kshitij Gupta
Qirui Sun
George Adamopoulos
Jonathan Siu Chi Lim
Quentin Gregory Anthony
Edwin Fennell
Irina Rish
The proliferation of Vision-Language Models (VLMs) in the past several years calls for rigorous and comprehensive evaluation methods and ben… (see more)chmarks. This work analyzes existing VLM evaluation techniques, including automated metrics, AI-based assessments, and human evaluations across diverse tasks. We first introduce Robin - a novel suite of VLMs that we built by combining Large Language Models (LLMs) and Vision Encoders (VEs) at multiple scales, and use Robin to identify shortcomings of current evaluation approaches across scales. Next, to overcome the identified limitations, we introduce CHIRP - a new long form response benchmark we developed for more robust and complete VLM evaluation. We provide open access to the Robin training code, model suite, and CHIRP benchmark to promote reproducibility and advance VLM research.
2025-01-16
ArXiv (preprint)
doi.org
arxiv.org
AfriHate: A Multilingual Collection of Hate Speech and Abusive Language Datasets for African Languages
Shamsuddeen Hassan Muhammad
Idris Abdulmumin
Abinew Ayele
David Ifeoluwa Adelani
Ibrahim Ahmad
Saminu Mohammad Aliyu
Nelson Odhiambo Onyango
Lilian D. A. Wanzare
Samuel Rutunda
Lukman Jibril Aliyu
Esubalew Alemneh
Oumaima Hourrane
Hagos Gebremichael
Elyas Abdi Ismail
Meriem Beloucif
Ebrahim Chekol Jibril
Andiswa Bukula
Rooweither Mabuya
Salomey Osei
Abigail Oppong … (see 7 more)
Tadesse Belay
Tadesse Kebede Guge
Tesfa Tegegne Asfaw
Chiamaka Ijeoma Chukwuneke
Paul Rottger
Seid Muhie Yimam
Nedjma OUSIDHOUM
Hate speech and abusive language are global phenomena that need socio-cultural background knowledge to be understood, identified, and modera… (see more)ted. However, in many regions of the Global South, there have been several documented occurrences of (1) absence of moderation and (2) censorship due to the reliance on keyword spotting out of context. Further, high-profile individuals have frequently been at the center of the moderation process, while large and targeted hate speech campaigns against minorities have been overlooked. These limitations are mainly due to the lack of high-quality data in the local languages and the failure to include local communities in the collection, annotation, and moderation processes. To address this issue, we present AfriHate: a multilingual collection of hate speech and abusive language datasets in 15 African languages. Each instance in AfriHate is annotated by native speakers familiar with the local culture. We report the challenges related to the construction of the datasets and present various classification baseline results with and without using LLMs. The datasets, individual annotations, and hate speech and offensive language lexicons are available on https://github.com/AfriHate/AfriHate
2025-01-14
ArXiv (preprint)
doi.org
arxiv.org
AfriHate: A Multilingual Collection of Hate Speech and Abusive Language Datasets for African Languages
Shamsuddeen Hassan Muhammad
Idris Abdulmumin
Abinew Ayele
David Ifeoluwa Adelani
Ibrahim Ahmad
Saminu Mohammad Aliyu
Nelson Odhiambo Onyango
Lilian D. A. Wanzare
Samuel Rutunda
Lukman Jibril Aliyu
Esubalew Alemneh
Oumaima Hourrane
Hagos Gebremichael
Elyas Abdi Ismail
Meriem Beloucif
Ebrahim Chekol Jibril
Andiswa Bukula
Rooweither Mabuya
Salomey Osei
Abigail Oppong … (see 7 more)
Tadesse Belay
Tadesse Kebede Guge
Tesfa Tegegne Asfaw
Chiamaka Ijeoma Chukwuneke
Paul Rottger
Seid Muhie Yimam
Nedjma OUSIDHOUM
Hate speech and abusive language are global phenomena that need socio-cultural background knowledge to be understood, identified, and modera… (see more)ted. However, in many regions of the Global South, there have been several documented occurrences of (1) absence of moderation and (2) censorship due to the reliance on keyword spotting out of context. Further, high-profile individuals have frequently been at the center of the moderation process, while large and targeted hate speech campaigns against minorities have been overlooked. These limitations are mainly due to the lack of high-quality data in the local languages and the failure to include local communities in the collection, annotation, and moderation processes. To address this issue, we present AfriHate: a multilingual collection of hate speech and abusive language datasets in 15 African languages. Each instance in AfriHate is annotated by native speakers familiar with the local culture. We report the challenges related to the construction of the datasets and present various classification baseline results with and without using LLMs. The datasets, individual annotations, and hate speech and offensive language lexicons are available on https://github.com/AfriHate/AfriHate
2025-01-14
ArXiv (preprint)
arxiv.org
arxiv.org
EarthView: A Large Scale Remote Sensing Dataset for Self-Supervision
Diego Velazquez
Pau Rodriguez
Sergio Alonso
Josep M. Gonfaus
Jordi Gonzàlez 0001
Gerardo Richarte
Javier Marin
Yoshua Bengio
Alexandre Lacoste
This paper presents EarthView, a comprehensive dataset specifically designed for self-supervision on remote sensing data, intended to enhanc… (see more)e deep learning applications on Earth monitoring tasks. The dataset spans 15 tera pixels of global remote-sensing data, combining imagery from a diverse range of sources, including NEON, Sentinel, and a novel release of 1m spatial resolution data from Satellogic. Our dataset provides a wide spectrum of image data with varying resolutions, harnessed from different sensors and organized coherently into an accessible HuggingFace dataset in parquet format. This data spans five years, from 2017 to 2022. Accompanying the dataset, we introduce EarthMAE, a tailored Masked Autoencoder, developed to tackle the distinct challenges of remote sensing data. Trained in a self-supervised fashion, EarthMAE effectively processes different data modalities such as hyperspectral, multispectral, topographical data, segmentation maps, and temporal structure. This model helps us show that pre-training on Satellogic data improves performance on downstream tasks. While there is still a gap to fill in MAE for heterogeneous data, we regard this innovative combination of an expansive, diverse dataset and a versatile model adapted for self-supervised learning as a stride forward in deep learning for Earth monitoring.
2025-01-14
ArXiv (preprint)
arxiv.org
arxiv.org
Integrating food webs in species distribution models can improve ecological niche estimation and predictions
Giovanni Poggiato
Jérémy Andréoletti
Laura J. Pollock
Wilfried Thuiller
2025-01-14
Ecography (published)
doi.org
Integrating food webs in species distribution models can improve ecological niche estimation and predictions
Giovanni Poggiato
Jérémy Andréoletti
Laura J. Pollock
Wilfried Thuiller
Biotic interactions play a fundamental role in shaping multitrophic species communities, yet incorporating these interactions into species d… (see more)istribution models (SDMs) remains challenging. With the growing availability of species interaction networks, it is now feasible to integrate these interactions into SDMs for more comprehensive predictions. Here, we propose a novel framework that combines trophic interaction networks with Bayesian structural equation models, enabling each species to be modeled based on its interactions with predators or prey alongside environmental factors. This framework addresses issues of multicollinearity and error propagation, making it possible to predict species distributions in unobserved locations or under future environmental conditions, even when prey or predator distributions are unknown. We tested and validated our framework on realistic simulated communities spanning different theoretical models and ecological setups. scenarios. Our approach significantly improved the estimation of both potential and realized niches compared to single SDMs, with mean performance gains of 8% and 6%, respectively. These improvements were especially notable for species strongly regulated by biotic factors, thereby enhancing model predictive accuracy. Our framework supports integration with various SDM extensions, such as occupancy and integrated models, offering flexibility and adaptability for future developments. While not a universal solution that consistently outperforms single SDMs, our approach provides a valuable new tool for modeling multitrophic community distributions when biotic interactions are known or assumed.
2025-01-14
Ecography (published)
doi.org