ISLAB/CAISR - User contributions [en]

Project Related to FeelAI -Collaboration with Volvo AMT

2025-10-27T05:10:21Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Time Series Forecasting with Incrementally Evolving Windows |TimeFrame=Fall 2025 |Supervisor=Zahra Taghiyarrenani, ? |Level=Master |Status=Op..."

{{StudentProjectTemplate
|Summary=Time Series Forecasting with Incrementally Evolving Windows
|TimeFrame=Fall 2025
|Supervisor=Zahra Taghiyarrenani, ?
|Level=Master
|Status=Open
}}
We study forecasting models that adapt as the available temporal context expands. Unlike fixed-window approaches, the model incrementally integrates new observations, refining its understanding of long-term trends and seasonal dynamics.

Zahra Taghiyarrenani

2025-10-27T05:01:00Z

Zahra:

zahra.taghiyarrenani@hh.se

https://www.hh.se/english/information-english/search-staff.html?person=3099FEE7-0238-41D8-B959-5056A3181737

Zahra Taghiyarrenani

2025-10-27T04:52:20Z

Zahra: Created page with "zahra.taghiyarrenani@hh.se"

zahra.taghiyarrenani@hh.se

Collaboration with Bankomat

2025-10-20T16:54:47Z

Zahra:

{{StudentProjectTemplate
|Summary=Explainable AI for Forecasting in Corporate Environments ( RQ: How to provide interpretable/explainable forecasts that align with business logic?
|TimeFrame=Fall2025
|Supervisor=Zahra Taghiyarrenani, Parisa?, Slawomir?, Sepideh?
|Level=Master
|Status=Open
}}
Explainable AI for Forecasting in Corporate Environments

We propose two master’s thesis projects in collaboration with Bankomat AB, focusing on the development of interpretable and explainable forecasting systems for ATM cash-demand prediction. Following, we describe how each of these thesis projects can be defined, representing two complementary directions.

Background and Motivation
Time-series forecasting plays a crucial role in financial and operational decision-making, where organizations must anticipate future demand, costs, or risks based on historical patterns. For an ATM network operator such as Bankomat AB, accurate cash-demand forecasting ensures that every machine has enough cash to meet customer needs while minimizing costly over-supply and transport inefficiencies. However, as forecasting models become increasingly data-driven and complex, they tend to operate as “black boxes.” This lack of transparency makes it difficult for decision-makers to understand why a model predicts higher withdrawals at a certain time or which factors drive a sudden change. Therefore, developing forecasting systems that are not only accurate but also explainable is essential for building trust, ensuring accountability, and aligning predictions with real-world business logic. Two promising directions for achieving this are:(1) Enhancing Transparency through Explainable AI in Forecasting, (2) leveraging Large Language Models for Contextual and Human-Centric Explanations.

Thesis1: Enhancing Transparency through Explainable AI in Forecasting
Explainable AI methods aim to uncover how input variables and temporal patterns influence forecasting outcomes [1, 2]. In the context of ATM cash-demand prediction, such methods can reveal which recent time window, seasonal pattern, or external variable (e.g., weekday, salary period, or temperature) most affects the forecast. Techniques such as SHAP, attention visualization, and prototype learning can quantify these effects and visualize the model’s reasoning, helping analysts trace how each decision is formed.
Research Question:
How can we design an efficient data-driven ATM cash-forecasting model that remains transparent by identifying and visualizing the most influential temporal and contextual factors behind each prediction?
This study will further explore how heterogeneous data sources, such as transaction history, holidays, temperature, and salary periods, can be integrated into an interpretable forecasting framework that combines predictive accuracy with human-understandable reasoning.

Thesis 2: Leveraging Large Language Models for Contextual and Human-Centric Explanations
Large Language Models (LLMs) offer a complementary approach to explainability by reasoning jointly over numerical and textual information [3, 4]. In ATM cash forecasting, LLMs can interpret external signals, such as news headlines, event reports, or policy announcements, to provide narrative explanations that connect real-world context with model predictions. Inspired by recent reflective forecasting frameworks [5], LLMs can dynamically integrate event information and refine their reasoning when prediction errors reveal missing contextual factors. This enables the generation of natural-language justifications, for example, explaining a forecasted increase in withdrawals as a result of a salary day coinciding with a regional festival.
Research Question:
How can LLMs be used to generate accurate and human-interpretable explanations for ATM cash-demand forecasts by incorporating both numerical data and external contextual information?
This study will further investigate how reflection-based reasoning mechanisms can improve the clarity, consistency, and business alignment of such explanations, supporting more transparent and trustworthy decision-making for Bankomat cash management operations.

Both theses aim to advance the transparency and practical usefulness of data-driven forecasting systems in financial operations. The XAI-based project is expected to deliver a forecasting framework that not only achieves high predictive accuracy but also provides clear visual and quantitative explanations for the factors influencing ATM cash demand. The LLM-based project, in turn, will develop a system that incorporates textual information, such as news, holidays, and regional events, to not only enhance forecasting accuracy but also generate meaningful insights into the underlying causes of demand fluctuations.

References:
1. Arsenault PD, Wang S, Patenaude JM. A survey of explainable artificial intelligence (XAI) in financial time series forecasting. ACM Computing Surveys. 2025 May 7;57(10):1-37.
2. Jahin MA, Shahriar A, Amin MA. MCDFN: supply chain demand forecasting via an explainable multi-channel data fusion network model. Evolutionary Intelligence. 2025 Jun;18(3):66.
3. Tang H, Zhang C, Jin M, Yu Q, Wang Z, Jin X, Zhang Y, Du M. Time series forecasting with llms: Understanding and enhancing model capabilities. ACM SIGKDD Explorations Newsletter. 2025 Jan 22;26(2):109-18.
4. Schoenegger P, Park PS, Karger E, Trott S, Tetlock PE. Ai-augmented predictions: Llm assistants improve human forecasting accuracy. ACM Transactions on Interactive Intelligent Systems. 2025 Feb 10;15(1):1-25.
5. Wang X, Feng M, Qiu J, Gu J, Zhao J. From news to forecast: Integrating event analysis in llm-based time series forecasting with reflection. Advances in Neural Information Processing Systems. 2024 Dec 16;37:58118-53.

Collaboration with Bankomat 5

2025-10-10T07:11:25Z

Zahra:

{{StudentProjectTemplate
|Summary=Forecasting with Sparse and Noisy Corporate Data RQ: What preprocessing techniques improve LLM performance on sparse transactional datasets?
|TimeFrame=Fall 2025
|Supervisor=Zahra Taghiyarrenani, TBD
|Level=Master
|Status=Open
}}

Collaboration with Bankomat 4

2025-10-10T07:10:53Z

Zahra:

{{StudentProjectTemplate
|Summary=Robustness of LLMs Against Prompt Injection Attacks RQ: What defense mechanisms can mitigate prompt injection vulnerabilities in enterprise LLM deployments?
|TimeFrame=Fall 2025
|Supervisor=Zahra Taghiyarrenani, TBD
|Level=Master
|Status=Open
}}

Collaboration with Bankomat 3

2025-10-10T07:10:32Z

Zahra:

{{StudentProjectTemplate
|Summary=LLM Feedback Loops for Autonomous Knowledge Updating RQ: What mechanisms allow LLMs to autonomously refine their knowledge based on user feedback?
|TimeFrame=Fall 2025
|Supervisor=Zahra Taghiyarrenani, TBD
|Level=Master
|Status=Open
}}

Collaboration with Bankomat 2

2025-10-10T07:10:00Z

Zahra:

{{StudentProjectTemplate
|Summary=Hybrid Models for Financial Forecasting Using LLMs and Time Series RQ: Can LLMs enhance traditional time-series models for forecasting financial KPIs?
|TimeFrame=Fall 2025
|Supervisor=Zahra Taghiyarrenani, Adeel, TBD
|Level=Master
|Status=Open
}}

Collaboration with Bankomat

2025-10-10T07:08:48Z

Zahra:

Collaboration with Bankomat

2025-10-10T07:07:30Z

Zahra:

Collaboration with Bankomat 5

2025-10-10T06:52:55Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Forecasting with Sparse and Noisy Corporate Data RQ: What preprocessing techniques improve LLM performance on sparse transactional datasets? ..."

Collaboration with Bankomat 4

2025-10-10T06:52:05Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Robustness of LLMs Against Prompt Injection Attacks RQ: What defense mechanisms can mitigate prompt injection vulnerabilities in enterprise L..."

Collaboration with Bankomat 3

2025-10-10T06:51:27Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=LLM Feedback Loops for Autonomous Knowledge Updating RQ: What mechanisms allow LLMs to autonomously refine their knowledge based on user feed..."

Collaboration with Bankomat 2

2025-10-10T06:50:22Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Hybrid Models for Financial Forecasting Using LLMs and Time Series RQ: Can LLMs enhance traditional time-series models for forecasting financ..."

Collaboration with Bankomat

2025-10-10T06:44:45Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Explainable AI for Forecasting in Corporate Environments ( RQ: How to provide interpretable/explainable forecasts that align with business lo..."

Utilization of Foundation Models for Federated Learning

2024-10-09T15:17:17Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=This thesis aims to leverage Foundation Models and develop new aggregation paradigms to overcome challenges in Federated Learning. |Superviso..."

{{StudentProjectTemplate
|Summary=This thesis aims to leverage Foundation Models and develop new aggregation paradigms to overcome challenges in Federated Learning.
|Supervisor=Zahra Taghiyarrenani, Yuantao Fan, Ali amirahmadi
|Level=Master
|Status=Open
}}
The rapid advancement of foundation models, such as large-scale language models and vision transformers, has revolutionized various fields in artificial intelligence (AI). These models, characterized by their immense scale and pre-training on vast datasets, offer impressive generalization capabilities across numerous tasks. Integrating these powerful models into federated learning (FL), a distributed machine learning approach presents a promising area for exploration.
The intersection of Federated Learning (FL) and Foundation Models (FM) introduces several potential benefits. For instance, FL can address challenges associated with non-iid (non-independent and identically distributed) and biased data by leveraging the advanced capabilities of FMs, leading to enhanced performance across various tasks and domains. More specifically, FMs can enhance FL in several ways:
FMs can provide a robust starting point for FL by offering pre-trained models that can be fine-tuned efficiently.
FMs can act as strong generators, synthesizing diverse data to enrich the training datasets used in FL.
FMs can serve as effective teachers through knowledge distillation, helping to address suboptimal performance in FL models.
Additionally, FMs introduce a new paradigm for sharing knowledge in FL. Unlike the traditional approach of exchanging high-dimensional model parameters, FMs enable the adoption of techniques like prompt tuning, offering a more efficient and flexible sharing mechanism within federated systems.
The aim of this thesis proposal is to explore the intersection of Federated Learning (FL) and Foundation Models (FM), with a primary focus on developing a novel aggregation method and a new paradigm for knowledge sharing among clients in FL.

Federated learning in automotive industry

2023-10-14T06:46:13Z

Zahra:

{{StudentProjectTemplate
|Summary=Federated learning in automotive industry
|Supervisor=Zahra Taghiyarrenani, Slawomir Nowaczyk, Sepideh Pashami
}}
With advancements in IoT and edge computing, the automotive industry needs to begin leveraging the benefits of Machine Learning in a federated setting. Federated Learning (FL) is an approach that allows various clients to collaboratively build Machine Learning (ML) models, transferring small amounts of information and ensuring privacy. As an example, the most well-known FL method is FedAvg, where all clients connect to a server. Every client trains a model and sends the model to the server. The server calculates the average of all models as a global model, and sends it back to clients. This procedure will continue until the convergence of the global model.

However, implementing a FL approach for the automotive industry faces many challenges. Some are related to the infrastructure -- for example, there are scenarios where sensors logging the data might be partially different, requiring Federated Transfer Learning. It is possible that the sensors are the same, but they operate on different schedules, requiring Asynchronous Federated Learning. It is also possible that the clients operate in different environments such that the factors influencing the decisions differ, which requires Heterogeneous Federated Learning. Moreover, depending on how much labeled and unlabelled data each client owns, one can consider supervised, semi-supervised, self-supervised, etc., setups.

As there are various types of challenges, the MSc thesis can start open-ended, and the topic(s) of most relevance for Federated Learning in the Automotive Industry can be explored.

The initial thesis plan:
1. A comprehensive analysis of the Challenges, Solutions, and Future of FL in the Automotive Industry. It is possible for a high-quality survey paper to be produced as a result of working on this topic.

2. Predictive maintenance can benefit the industry by minimizing the costs and risks associated with maintenance. Predictive maintenance models, such as the Remaining Useful Lifetime (RUL) predictor, are of interest in federated settings. Scientifically, this topic is related to Federated Learning for regression, specifically in heterogeneous situations (system heterogeneity or statistical heterogeneity), which is understudied.

3. Detecting anomalies is a critical task in predictive maintenance, safety, efficiency, etc. -- and, consequently, in the automotive industry. Thus, Federated Anomaly Detection is another potential area of study.

4. Data availability, specifically when it comes to solving real-world tasks, is a challenge. Semi-supervised and Self-supervised Federated Learning has shown promising results and should be explored in the automotive context.

Federated learning in automotive industry

2023-10-02T14:59:24Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=- |Supervisor=Zahra Taghiyarrenani, Slawomir Nowaczyk, Sepideh Pashami }} With advancements in IoT and edge computing, the automotive industr..."

{{StudentProjectTemplate
|Summary=-
|Supervisor=Zahra Taghiyarrenani, Slawomir Nowaczyk, Sepideh Pashami
}}
With advancements in IoT and edge computing, the automotive industry needs to begin leveraging the benefits of Machine Learning in a federated setting. Federated Learning (FL) is an approach that allows various clients to collaboratively build Machine Learning (ML) models, transferring small amounts of information and ensuring privacy. As an example, the most well-known FL method is FedAvg, where all clients connect to a server. Every client trains a model and sends the model to the server. The server calculates the average of all models as a global model, and sends it back to clients. This procedure will continue until the convergence of the global model.

However, implementing a FL approach for the automotive industry faces many challenges. Some are related to the infrastructure -- for example, there are scenarios where sensors logging the data might be partially different, requiring Federated Transfer Learning. It is possible that the sensors are the same, but they operate on different schedules, requiring Asynchronous Federated Learning. It is also possible that the clients operate in different environments such that the factors influencing the decisions differ, which requires Heterogeneous Federated Learning. Moreover, depending on how much labeled and unlabelled data each client owns, one can consider supervised, semi-supervised, self-supervised, etc., setups.

As there are various types of challenges, the MSc thesis can start open-ended, and the topic(s) of most relevance for Federated Learning in the Automotive Industry can be explored.

The initial thesis plan:
1. A comprehensive analysis of the Challenges, Solutions, and Future of FL in the Automotive Industry. It is possible for a high-quality survey paper to be produced as a result of working on this topic.

2. Predictive maintenance can benefit the industry by minimizing the costs and risks associated with maintenance. Predictive maintenance models, such as the Remaining Useful Lifetime (RUL) predictor, are of interest in federated settings. Scientifically, this topic is related to Federated Learning for regression, specifically in heterogeneous situations (system heterogeneity or statistical heterogeneity), which is understudied.

3. Detecting anomalies is a critical task in predictive maintenance, safety, efficiency, etc. -- and, consequently, in the automotive industry. Thus, Federated Anomaly Detection is another potential area of study.

4. Data availability, specifically when it comes to solving real-world tasks, is a challenge. Semi-supervised and Self-supervised Federated Learning has shown promising results and should be explored in the automotive context.

Transfer Learning for Network Security

2020-10-11T10:58:41Z

Zahra: Created page with "{{StudentProjectTemplate |Summary=Study of Transfer Learning techniques in Network Security applications- Network Traffic Classification and Intrusion Detection |Keywords=Tran..."

{{StudentProjectTemplate
|Summary=Study of Transfer Learning techniques in Network Security applications- Network Traffic Classification and Intrusion Detection
|Keywords=Transfer Learning, Domain Adaptation, Traffic Classification, Intrusion Detection
|References=Li D, Yuan Q, Li T, Chen S, Yang J. Cross-domain Network Traffic Classification Using Unsupervised Domain Adaptation. In2020 International Conference on Information Networking (ICOIN) 2020 Jan 7 (pp. 245-250). IEEE.

Sun G, Liang L, Chen T, Xiao F, Lang F. Network traffic classification based on transfer learning. Computers & electrical engineering. 2018 Jul 1;69:920-7.

Taghiyarrenani Z, Fanian A, Mahdavi E, Mirzaei A, Farsi H. Transfer learning based intrusion detection. In2018 8th International Conference on Computer and Knowledge Engineering (ICCKE) 2018 Oct 25 (pp. 92-97). IEEE.
|Supervisor=Slawomir Nowaczyk, Zahra Taghiyarrenani
|Level=Master
|Status=Open
}}
These days, utilization of Transfer Learning and more specifically Domain Adaptation is increasingly getting researchers' attention specifically for solving real-world problems. Network security applications including intrusion detection and network traffic classification can also make a profit from this technique.

As a general definition, transfer learning methods extract knowledge from one domain(Source) and employ them for solving the problem in another domain(Target). The definition of the domain depends on the problem to be solved. Particularly, in the network field, we can refer to each separate network as a domain. So, using Transfer Learning, it would be possible to use the available(labeled) samples from one network to train a learning model in another network. Despite the fact that there are many developed transfer learning methods, the utilization of that in the network security fields is not investigated enough yet.

The main objective of this work is 1)to study the challenges behind the use of the Transfer learning in network security applications 2) to study the different transfer learning techniques including instance_based, feature_based, and model_based methods that can be applied in this field, and finally 3) develop a new transfer learning method for this field.