tag:blogger.com,1999:blog-52843700126859613842024-03-12T22:20:36.399+00:00Alejandro Mosquera | BlogNon-stochastic musings about Computer Science, NLP, Cybersecurity and MLAlejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.comBlogger10125tag:blogger.com,1999:blog-5284370012685961384.post-18070151259681465762024-02-17T15:50:00.001+00:002024-02-17T15:52:12.615+00:00Detecting LLM hallucinations and overgeneration mistakes @ SemEval 2024 The modern NLG landscape is plagued by two interlinked problems: On the one hand, our current neural models have a propensity to produce inaccurate but fluent outputs; on the other hand, our metrics are most apt at describing fluency, rather than correctness. This leads neural networks to “hallucinate”, e.g., produce fluent but incorrect outputs that we currently struggle to detect Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-41562244832855157202023-06-05T07:15:00.010+00:002023-06-05T07:27:16.873+00:00Hackaprompt-2023 @ AICrowd write upHackAPrompt was a prompt hacking competition aimed at enhancing AI safety and education by challenging participants to outsmart large language models (e.g. ChatGPT, GPT-3). In particular, encouraged participants to attempt to hack through many prompt hacking defenses as possible.The task organizers provided a set of 10 challenges in which the participants should bypass Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-14184695716209310842023-05-13T18:59:00.004+00:002023-05-15T06:56:19.536+00:00Living off the land: Solving ML problems without training a single modelIntroductionThe concept of living off the land is related to surviving on what you can forage, hunt, or grow in nature.Considering the current Machine Learning landscape, we can draw a parallelism between living off the land and "shopping around" for ready-made models for a given task. While this has been partially true for some time thanks to model repositories such as HuggingFace, it Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-87018379514609627802023-02-16T21:51:00.004+00:002023-03-12T13:21:55.367+00:00Pretrained Models with Adversarial Training for Online Sexism Detection @ SemEval 2023 Abstract Adversarial training can provide neural networks with significantly improved resistance to adversarial attacks, thus improving model robustness. However, a major drawback of many existing adversarial training workflows is the computational cost and extra processing time when using data augmentation techniques. This post explores the Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-5345496386261206142023-01-24T20:00:00.001+00:002023-03-12T13:22:35.130+00:00 The string similarity problemFor two strings A and B (in the ASCII [a-z] range), we define the similarity of the strings to be the length of the longest prefix common to both strings. For example, the similarity of strings "abc" and "abd" is 2, while the similarity of strings "aaa" and "aaab" is 3.The reader is asked to calculate the sum of similarities of a string S with each of its suffixes. Reference (https://Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-50969564544923277232022-12-07T08:31:00.004+00:002023-03-12T13:23:03.046+00:00Revisiting the Microsoft Malware Classification Challenge (BIG 2015) in 2022 In 2015, Microsoft provided the data science community with an unprecedented malware dataset and encouraging open-source progress on effective techniques for grouping variants of malware files into their respective families. Formatted as a Kaggle Competition, it featured a very large (for that time) dataset comprising of almost 40GB of compressed files containing disarmed malware samples Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-45046831705703185192022-12-06T21:37:00.001+00:002023-03-12T13:23:28.983+00:00On the Intriguing Properties of Backdoored Neural NetworksIntroduction Malicious actors can alter the expected behavior of a neural network in order to respond to data containing certain triggers only known to the attacker, without disrupting model performance when presented with normal inputs. An adversary will commonly force these misclassifications by either performing trigger injection [19] or dataset poisoning [6]. Less popular techniques thatAlejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-86297254349009780712022-09-27T08:37:00.008+00:002023-03-12T13:23:49.155+00:00AI Village Capture the Flag @ DEFCON write up In August 2022 I had the chance to participate in an AI-themed CTF collocated with the DEF CON 30 security (hacking) conference. This was particularly interesting since it was presented in a novel format as a Kaggle competition where the leaderboard was ranked based on the points that each of the discovered flags was providing. Despite entering the competition in its latest stage I did Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-69692413080732728542022-03-17T20:18:00.004+00:002023-03-12T13:24:04.730+00:00Defending and attacking ML Malware Classifiers for Fun and Profit: 2x prize winner at MLSEC-2021MLSEC (Machine Learning Security Evasion Competition) is an initiative sponsored by Microsoft and partners CUJO AI, NVIDIA, VMRay, and MRG Effitas with the purpose of raising awareness of the expanding attack surface which is now also affecting AI-powered systems. In its 3rd edition the competition allowed defenders and attackers to exercise their security and machine learning skills Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0tag:blogger.com,1999:blog-5284370012685961384.post-64785159019319722192022-03-17T15:28:00.007+00:002023-03-12T13:24:23.156+00:00Towards Machines that Capture and Reason with Science Knowledge In 2015 I took part on a machine learning competition hosted on Kaggle aiming to solve a multiple-question 8th grade science test. At that time there weren't large pretrained models to leverage and (unsurprisingly) best performing models were IR-based that would barely achieve a GPA of 1.0 in the US grading system:However, several years later (and several thousands of $$$ spent Alejandrohttp://www.blogger.com/profile/09513085903438138402noreply@blogger.com0