A5021932110- Adversarial Robustness in Machine Learning
- Electric Vehicles and Infrastructure
- Vehicle emissions and performance
- Topic Modeling
- Energy, Environment, and Transportation Policies
- Botany and Plant Ecology Studies
- Electric Power Systems and Control
- Soil Mechanics and Vehicle Dynamics
- Hate Speech and Cyberbullying Detection
- Domain Adaptation and Few-Shot Learning
- Forest ecology and management
- Anomaly Detection Techniques and Applications
- Advanced Battery Technologies Research
- Advanced Neural Network Applications
- Genetic diversity and population structure
- Agricultural Engineering and Mechanization
- Natural Language Processing Techniques
- Agriculture and Farm Safety
- Explainable Artificial Intelligence (XAI)
- Multimodal Machine Learning Applications
- Diverse Industrial Engineering Technologies
- Network Security and Intrusion Detection
- Mediterranean and Iberian flora and fauna
- Maritime Transport Emissions and Efficiency
- Misinformation and Its Impacts
Angel Kanchev University of Ruse
2017-2024
Art Institute of Portland
2023
University of Waterloo
2022
University of Forestry
2015-2018
New Bulgarian University
2013-2015
We release Code Llama, a family of large language models for code based on Llama 2 providing state-of-the-art performance among open models, infilling capabilities, support input contexts, and zero-shot instruction following ability programming tasks. provide multiple flavors to cover wide range applications: foundation (Code Llama), Python specializations - Python), instruction-following Instruct) with 7B, 13B, 34B 70B parameters each. All are trained sequences 16k tokens show improvements...
Machine learning models have been found to learn shortcuts—unintended decision rules that are unable generalize—undermining models' reliability. Previous works address this problem under the tenuous assumption only a single shortcut exists in training data. Real-world images rife with multiple visual cues from background texture. Key advancing reliability of vision systems is understanding whether existing methods can overcome shortcuts or struggle Whac-A-Mole game, i.e., where mitigating...
The paper analyzes the power demand of auxiliary systems electric cars. On basis existing cars an analysis energy consumption different is done. As a result possibilities for rational use these have been proposed, which can increase mileage per one charge battery.
Large-scale automatic speech translation systems today lack key features that help machine-mediated communication feel seamless when compared to human-to-human dialogue. In this work, we introduce a family of models enable end-to-end expressive and multilingual translations in streaming fashion. First, contribute an improved version the massively multimodal SeamlessM4T model-SeamlessM4T v2. This newer model, incorporating updated UnitY2 framework, was trained on more low-resource language...
The genus Abies is represented in southern Balkans by A. alba, cephalonica, and borisii-regis. To infer the status of southern-Balkans firs, as well extent patterns introgression within this taxonomical complex, we analyzed genetic variation 29 indigenous fir populations Bulgaria, Macedonia, Greece, Calabria using a combination maternally biparentally inherited markers. Three mitochondrial lineages were observed, one comprising Calabrian two distributed Balkans, coinciding with alba...
This paper presents CyberSecEval, a comprehensive benchmark developed to help bolster the cybersecurity of Large Language Models (LLMs) employed as coding assistants. As what we believe be most extensive unified safety date, CyberSecEval provides thorough evaluation LLMs in two crucial security domains: their propensity generate insecure code and level compliance when asked assist cyberattacks. Through case study involving seven models from Llama 2, Code Llama, OpenAI GPT large language...
Deep learning vision systems are widely deployed across applications where reliability is critical. However, even today's best models can fail to recognize an object when its pose, lighting, or background varies. While existing benchmarks surface examples challenging for models, they do not explain why such mistakes arise. To address this need, we introduce ImageNet-X, a set of sixteen human annotations factors as background, lighting the entire ImageNet-1k validation well random subset 12k...
The paper presents a construction of an experimental electric bicycle for evaluation the energy efficiency. is equipped with onboard computer which can store information about motion and consumption. result concerning power, consumption, recharging during brake process, etc. are given. Energy consumption 3 typical city routes studied.
Textual prompt tuning has demonstrated significant performance improvements in adapting natural language processing models to a variety of downstream tasks by treating hand-engineered prompts as trainable parameters. Inspired the success textual prompting, several studies have investigated efficacy visual tuning. In this work, we present Visual Prompt Adaptation (VPA), first framework that generalizes prompting with test-time adaptation. VPA introduces small number learnable tokens, enabling...
The paper presents an upgraded methodology for determination of the electric motor power considering time acceleration. influence speed factor on value needed at same acceleration is studied. Some calculations basis real vehicle were made. numeric and graphical results are given. They show a decrease with increase motor, because high allows use larger range characteristic maximum motor. An experimental verification was done.
This paper presents an analysis concerning the effectiveness of electric traction in comparison with conventional cars. The Life Cycle Assessment method is used. It estimates energy spent for extraction raw materials/sources, manufacturing and transportation components vehicle, motion, maintenance repair during exploitation period recycling process. impact production technology energy, needed charging battery, taken into account. consumption CO 2 emissions life cycle cars are presented on...
A major barrier towards the practical deployment of large language models (LLMs) is their lack reliability. Three situations where this particularly apparent are correctness, hallucinations when given unanswerable questions, and safety. In all three cases, should ideally abstain from responding, much like humans, whose ability to understand uncertainty makes us refrain answering questions we don't know. Inspired by analogous approaches in classification, study explores feasibility efficacy...
In recent years, regarding the influence of production processes and vehicles on environment, new technical solutions for reducing...
In the paper, an evaluation of ecological effect electric vehicles, using energy, produced from different types power stations is done. On base Emission factor (CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> g/kWh) energy production renewable sources and fossil fuels, existing life cycle standards vehicles The conventional gasoline are used as a for comparison. Using emission factors EC-28 countries, mathematical models concerning...
This work examines the vulnerability of multimodal (image + text) models to adversarial threats similar those discussed in previous literature on unimodal (image- or text-only) models. We introduce realistic assumptions partial model knowledge and access, discuss how these differ from standard "black-box"/"white-box" dichotomy common current attacks. Working under various levels "gray-box" assumptions, we develop new attack methodologies unique classification evaluate them Hateful Memes...
The paper presents results of a study on progeny test in Yundola established 1966 and comprising 26 local plus trees Scots Pine. Diameters at breast height (DBH) were measured 1996 2012. There substantial differences among the performance half-sib families regarding diameter growth. best performing 30 46 years after planting same, but there statistically significant changes rank as whole. Mean survival rate was 91% age 79% 46, which is relatively high. Individual heritability increased from...
Assessing performance in Natural Language Processing is becoming increasingly complex. One particular challenge the potential for evaluation datasets to overlap with training data, either directly or indirectly, which can lead skewed results and overestimation of model performance. As a consequence, human gaining increasing interest as means assess reliability models. such method red teaming approach, aims generate edge cases where will produce critical errors. While this methodology...
Genetic diversity of eleven natural populations Silver Fir in Bulgaria was studied by means allozyme gene markers. Within population varied among and generally high, but within the range reported for other European species. Mean number alleles per locus ranged from 2.1 to 2.4 effective allele &ndash; 1.2 1.5. Expected observed heterozygo-sities were also high (mean values 0.295 0.265, respectively), there significant heterozygote deficiency most loci populations, resulting positive...
In this research, the energy consumption and emissions of gasoline cars, battery electric fuel cells (FCE) compressed naturel gas (CNG) liquid propane-butane (LPG) etc., during their life cycles are presented. A set flow charts presenting production cars used fuel/energy is developed. These latest data for electricity emission factor raw materials prices used. comparative analysis done. Results primary carbon presented in graphical form. They show most economical ecological car decisions.
In the presented paper flow charts for Life cycle of main types fuels/energy sources and cars is done. Respective mathematical models, taking into account stages material production, exploitation recycling are developed. Based on energy production 2015 2020 years mix European countries Bulgaria, an assessment consumption CO2 emissions (LCA) using gasoline, natural gas, hydrogen fuel cells etc. show advantages disadvantages different type vehicles, during its entire cycle. Development in...
Red teaming assesses how large language models (LLMs) can produce content that violates norms, policies, and rules set during their safety training. However, most existing automated methods in the literature are not representative of way humans tend to interact with AI models. Common users may have advanced knowledge adversarial machine learning or access model internals, they do spend a lot time crafting single highly effective prompt. Instead, likely make use techniques commonly shared...
Micromeria juliana is a rare species in Bulgaria and also valuable medicinal plant. It included the Red Data Book of with conservation status “endangered” protected by Biodiversity Act Bulgaria. The present report focuses to study natural populations plant communities M. two regions its occurrence – Eastern Rhodopes Valley Mesta River. In first locality grows on steep stony slopes walls medieval fortress, second one calcareous slope, abandoned agricultural land. We results survey composition...