Unicode · Bulletin of Miscellaneous Information (Royal Gardens Kew)
"Unicode" is the self-titled first release by this new tonstube artist. This short five track EP is his most personal, experimental and dark work to date. Being produced completly in linux it also is the starting point of a new production aproach for all future tonstube releases. With elements from idm, breakbeat, glitch and experimental, unicode is a statement against the polished, generic and over produced releases of today's music industry.
Cited by 0
Brčić, Filip · Hispana · 2003
We review coding of multi-language text in digital form using Unicode standard, with special attention to UTF-8 variant, which is the most convenient variant for coding latin text. We also give a short tutorial for using UTF-8 in Microsoft Word, Netscape Composer and text editor Kate. Standard Unicode fonts are recommended so that the texts can be easily transfered from a computer to another one or for publishing on Internet.
Cited by 0Open access
??? · The Royal Society of Chemistry’s Journals, Books and Databases (The Royal Society of Chemistry) · 2015
Cited by 0
Yannis Haralambous · 2024
Cited by 1
Brčić, Filip · E-LIS Repository (University of Naples Federico II) · 2003
We review coding of multi-language text in digital form using Unicode standard, with special attention to\nUTF-8 variant, which is the most convenient variant for coding latin text. We also give a short tutorial for\nusing UTF-8 in Microsoft Word, Netscape Composer and text editor Kate. Standard Unicode fonts are\nrecommended so that the texts can be easily transfered from a computer to another one or for publishing\non Internet.
Cited by 0Open access
Kurt Hornik · 2010
Cited by 2Open access
M. Davis, L. Collins · 2002
The results of the Unicode project, an effort to improve the architecture of multilingual text handling on the Macintosh, are reported. A method of encoding characters on computers that allows for efficient processing with a character set that covers all the world languages is presented. The discussion covers the fundamental design principles for Unicode, the criteria for distinguishing letters, the contrasts to an alternative coding scheme (ISO draft 10646), and operating system requirements for effective implementation.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">></ETX>
Cited by 5
Ethan V. Munson · 2018
Cited by 0
songrui li · Zenodo (CERN European Organization for Nuclear Research) · 2025
ContentsThis repository provides the official implementation of the paper 'Enhancing Code Model Robustness Against Identifier Renaming via Unified Code Normalization'. DescriptionDeep code models (DCMs) are increasingly deployed in security-critical applications. Yet, their vulnerability to adversarial perturbations – such as subtle identifier renaming – poses significant risks, as these changes can induce out-of-distribution inputs and cause insecure predictions. A key challenge lies in defending against such attacks without prior knowledge of adversarial patterns, as the space of possible perturbations is virtually infinite, and conventional rule-based defenses fail to generalize. To address this challenge, we primarily focus on defending renaming-based adversarial attacks, which have the most significant impact on DCMs’ security, and propose a novel two-stage defense framework named UniCode, which proactively normalizes adversarial inputs into uniformly in-distribution representations. Please refer to overview.jpg for a detailed overview of our method's structure. Specifically, the first stage strips all identifiers into placeholders, eliminating adversarial influence while maintaining the code structure and functionality, and the second stage reconstructs semantically meaningful identifiers by leveraging contextual understanding from large language models, ensuring the comprehensive code semantics are preserved. By fine-tuning the code models on the normalized distribution, UniCode renders models inherently robust against diverse renaming attacks without requiring attack-specific adaptations. To evaluate the performance of our approach, we have conducted a comprehensive evaluation by comparing it with state-of-the-art baseline methods. The experimental results demonstrate that UniCode achieves the best defense performance on 82.22% of subjects, with average improvements ranging from 9.86% to 46.1% over baselines in terms of defending effectiveness, indicating the superior performance of UniCode.StructureHere, we briefly introduce the usage of each directory in UNICODE:```├─ALERT (Our baseline, for each baseline we provide code on three model/datasets due to space limited)│ ├─codebert│ ├─codet5│ ├─graphcodebert├─CDenoise (Our baseline, CodeDenoise)├─CODA (Our baseline)├─ITGen (Our baseline)├─MARVEL (Our baseline)├─CodeTAE (Our baseline)│─code (Our Approach, UniCode)│ ├─abstract.py (abstracting identifier names)│ ├─replace_method_name.py (abstracting function names)│ ├─normalization.py (conducting code instantiation using LLM)│ ├─build_txt.py (conducting data pre-processing)│ ├─VulnerabilityPrediction_build_jsonl.py (conducting data pre-processing)├─python_parser (parsing code for further analysis)```Datasets/ModelsThe overall model and datasets used in this paper has been listed below. All experimental datasets are provided in data.zip, and the corresponding model weights are stored in the current directory.| Task | Dateset | Train/Val/Test | acc || :----------------------: | :--------: | :----------------: | :----: || Clone Detection | CodeBERT | 90,102/4,000/4,000 | 96.88% || | GCBERT | | 96.73% || | CodeT5 | | 96.40% || | CodeT5Plus | | 97.47% || Vulnerability Prediction | CodeBERT | 21,854/2,732/2,732 | 63.76% || | GCBERT | | 64.13% || | CodeT5 | | 59.99% || | CodeT5Plus | | 58.13% || Defect Prediction | CodeBERT | 27,058/–/6,764 | 84.37% || | GCBERT | | 84.89% || | CodeT5 | | 88.82% || | CodeT5Plus | | 88.99% |Requirements:- python==3.7.7- transformers==4.8.2- pytorch==1.5.1- pandas==1.3.0ReproducibilityUsage InstructionsTo leverage our abstract framework and instantiate methods, follow these steps:Configure Paths Modify the following variables in the code:```input_path = "your/input/path" # Replace with your input directoryoutput_path = "your/output/path" # Specify desired output location```API Key Setup Replace the placeholder with your LLM provider's API key:```api_key = "your_api_key_here" # e.g.deepseek,gpt4-o-mini```Execution Run the core pipeline with:```cd code;python abstract.py --dataset <datasetname> --model <datasetname>python normalization.py --dataset <datasetname> --model <datasetname>````--dataset` refers to the dataset used for evaluation`--model` refers to the model used for evaluation
Cited by 0Open access
Ethan V. Munson · Encyclopedia of Database Systems · 2018
Cited by 0
Apress eBooks · 2009
If you’ve ever come across text in a foreign language that contains lots of question mark characters in unexpected positions or if you’ve written Python code that causes an exception such as the following one to be raised, then chances are you have run into a problem with character sets, encodings, and Unicode: UnicodeDecodeError: ‘ascii’ codec can’t decode byte 0xff in position 6:➥ ordinal not in range(128)
Cited by 0
Georg Ch. Pflug · Lexikon des gesamten Buchwesens Online · 2020
Cited by 1
Ed Pinto · Bulletin of Miscellaneous Information (Royal Gardens Kew)
Cited by 0
SpringerReference · 2011
Cited by 0
Ethan V. Munson · Encyclopedia of Database Systems · 2009
Cited by 1
brian d foy, Thomas Sibley, Peter Mayr et al. · Zenodo (CERN European Organization for Nuclear Research) · 2025
1.118 2025-01-03T06:22:33Z refresh distro and move to BRIANDFOY
Cited by 0Open access
brian d foy · Zenodo (CERN European Organization for Nuclear Research) · 2025
1.036 2025-01-03T02:58:30Z refresh distro move to BRIANDFOY user on PAUSE add security policy
Cited by 0Open access
Joe Celko · Elsevier eBooks · 2009
Cited by 0
songrui li · Zenodo (CERN European Organization for Nuclear Research) · 2025
ContentsThis repository provides the official implementation of the paper 'Enhancing Code Model Robustness Against Identifier Renaming via Unified Code Normalization'. DescriptionDeep code models (DCMs) are increasingly deployed in security-critical applications. Yet, their vulnerability to adversarial perturbations – such as subtle identifier renaming – poses significant risks, as these changes can induce out-of-distribution inputs and cause insecure predictions. A key challenge lies in defending against such attacks without prior knowledge of adversarial patterns, as the space of possible perturbations is virtually infinite, and conventional rule-based defenses fail to generalize. To address this challenge, we primarily focus on defending renaming-based adversarial attacks, which have the most significant impact on DCMs’ security, and propose a novel two-stage defense framework named UniCode, which proactively normalizes adversarial inputs into uniformly in-distribution representations. Please refer to overview.jpg for a detailed overview of our method's structure. Specifically, the first stage strips all identifiers into placeholders, eliminating adversarial influence while maintaining the code structure and functionality, and the second stage reconstructs semantically meaningful identifiers by leveraging contextual understanding from large language models, ensuring the comprehensive code semantics are preserved. By fine-tuning the code models on the normalized distribution, UniCode renders models inherently robust against diverse renaming attacks without requiring attack-specific adaptations. To evaluate the performance of our approach, we have conducted a comprehensive evaluation by comparing it with state-of-the-art baseline methods. The experimental results demonstrate that UniCode achieves the best defense performance on 82.22% of subjects, with average improvements ranging from 9.86% to 46.1% over baselines in terms of defending effectiveness, indicating the superior performance of UniCode.StructureHere, we briefly introduce the usage of each directory in UNICODE:```├─ALERT (Our baseline, for each baseline we provide code on three model/datasets due to space limited)│ ├─codebert│ ├─codet5│ ├─graphcodebert├─CDenoise (Our baseline, CodeDenoise)├─CODA (Our baseline)├─ITGen (Our baseline)├─MARVEL (Our baseline)├─CodeTAE (Our baseline)│─code (Our Approach, UniCode)│ ├─abstract.py (abstracting identifier names)│ ├─replace_method_name.py (abstracting function names)│ ├─normalization.py (conducting code instantiation using LLM)│ ├─build_txt.py (conducting data pre-processing)│ ├─VulnerabilityPrediction_build_jsonl.py (conducting data pre-processing)├─python_parser (parsing code for further analysis)```Datasets/ModelsThe overall model and datasets used in this paper has been listed below. All experimental datasets are provided in data.zip, and the corresponding model weights are stored in the current directory.| Task | Dateset | Train/Val/Test | acc || :----------------------: | :--------: | :----------------: | :----: || Clone Detection | CodeBERT | 90,102/4,000/4,000 | 96.88% || | GCBERT | | 96.73% || | CodeT5 | | 96.40% || | CodeT5Plus | | 97.47% || Vulnerability Prediction | CodeBERT | 21,854/2,732/2,732 | 63.76% || | GCBERT | | 64.13% || | CodeT5 | | 59.99% || | CodeT5Plus | | 58.13% || Defect Prediction | CodeBERT | 27,058/–/6,764 | 84.37% || | GCBERT | | 84.89% || | CodeT5 | | 88.82% || | CodeT5Plus | | 88.99% |Requirements:- python==3.7.7- transformers==4.8.2- pytorch==1.5.1- pandas==1.3.0ReproducibilityUsage InstructionsTo leverage our abstract framework and instantiate methods, follow these steps:Configure Paths Modify the following variables in the code:```input_path = "your/input/path" # Replace with your input directoryoutput_path = "your/output/path" # Specify desired output location```API Key Setup Replace the placeholder with your LLM provider's API key:```api_key = "your_api_key_here" # e.g.deepseek,gpt4-o-mini```Execution Run the core pipeline with:```cd code;python abstract.py --dataset <datasetname> --model <datasetname>python normalization.py --dataset <datasetname> --model <datasetname>````--dataset` refers to the dataset used for evaluation`--model` refers to the model used for evaluation
Cited by 0Open access
brian d foy, Thomas Sibley, Peter Mayr et al. · Zenodo (CERN European Organization for Nuclear Research) · 2025
1.122 2025-04-16T06:36:37Z require a newer version.pm for v5.10.1 tests
Cited by 0Open access