Data- och informationsteknik (CSE) // Computer Science and Engineering (CSE)

Använd denna länk för att länka till samlingen:

https://odr.chalmers.se/handle/20.500.12380/11

Vi utbildar för framtiden och skapar samhällsnytta genom vår forskning som levandegörs i nära samarbete med näringslivet. Vi bedriver forskning inom computer science, datateknik, software engineering och interaktionsdesign - från grundforskning till direkta tillämpningar. Institutionen har en stark internationell prägel och är delad mellan Chalmers och Göteborgs universitet.

Läs mer om våra utbildningar

För forskning och forskningspublikationer, se https://research.chalmers.se/organisation/data-och-informationsteknik/

We are engaged in research and education across the full spectrum of computer science, computer engineering, software engineering, and interaction design, from foundations to applications. We educate for the future, conduct research with high international visibility, and create societal benefits through close cooperation with businesses and industry. The department is joint between Chalmers and the University of Gothenburg.

Studying at the Department of Computer Science and Engineering at Chalmers

For research and research output, please visit https://research.chalmers.se/en/organization/computer-science-and-engineering/

Browse

Visar 1 - 20 av 37

A probabilistic model for genetic regulation of metabolic networks
(2013) Kallus, Jonatan; Wilsson, Joel; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Recent advancements in gene expression pro ling and measurement of metabolic reaction rates have led to increased interest in predicting metabolic reaction rates. In this thesis we present a principled approach for using gene expression pro les to improve predictions of metabolic reaction rates. A probabilistic graphical model is presented, which addresses inherent weaknesses in the current state of the art method for data-driven reconstruction of regulatory-metabolic networks. Our model combines methods from systems biology and machine learning, and is shown to outperform the current state of the art on synthetic data. Results on real data from S. cerevisiae and M. tuberculosis are also presented.
Accelerating Proximity Queries Accelerating Proximity Queries for Non-convex Geometries in a Robot Cell Context
(2018) Thorén, Joakim; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Sampling-based motion-planners, for example rapidly exploring dense tree (RRT) based planners, depend on fast proximity queries. Regrettably, bounding volume tests are significant bottlenecks of proximity queries. Sampling-based motion-planners are therefore accelerated by reducing the number of bounding volume tests. To this end, a novel algorithm called Forest Proximity Query (FPQ) is developed. Contrary to previous research, FPQ traverses several pairs of BVHs simultaneously, effectively exploiting an actuality that only a single minimal separation distance — out of several possible separation distances — is required during sampling-based motion-planning. An implementation of FPQ show that FPQ performs up to 67% fewer BV tests in comparison to the well-known Proximity Query Package, increasing proximity querying performance by up to 46%. In conclusion, FPQ is successful in its attempt at improving performance of sampling-based motion-planners.
Adaptive Bounding Volume Hierarchies For Deformable Surface Models
(2011) Bitar, Fadi; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
This master’s thesis explores a new mechanism for maintaining Bounding-Volume Hierarchies (BVH) of deformable surface models. Typical algorithms found in the literature are based on refitting only a portion of the BVH, leaving sometimes a large portion of the Bounding Volumes (BVs) inaccurately representing the parts of the object they should enclose. The algorithm proposed in this thesis allows the BVH’s quality to degrade as the model it represents deforms, while guaranteeing that every point in the model is contained within the BVH at all times, and thus maintaining the accuracy of any collision detection or distance measurement queries performed on the model. Through a tunable asynchronous refitting of the individual bounding volumes, the algorithm offers a computationally efficient, low memory cost solution to the accurate simulation of deformable surface models in real environments. The decision criteria for the refitting of the BVs along with the parameters of these criteria are optimized through a Genetic Algorithm search. The resulting algorithm is shown to outperform the most commonly referred to BVH-based algorithm referred to in the literature.
An Incompressible Navier-Stokes Equations Solver on the GPU Using CUDA
(2013) Karlsson, Niklas; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Graphics Processing Units (GPUs) have emerged as highly capable computational accelerators for scientific and engineering applications. Many reports claim orders of magnitude of speedup compared to traditional Central Processing Units (CPUs), and the interest for GPU computation is high in the computational world. In this thesis, the capability of using GPUs to accelerate the full computational chain of a 3D incompressible Navier-Stokes solver, including solvers and preconditioners for sparse linear systems as well as assembly routines for a finite volume discretization, has been evaluated. The CG, GMRES and BiCGStab iterative solvers have been implemented on the CUDA GPGPU platform and evaluated together with the Jacobi, and Least Square Polynomial preconditioners. A double precision Navier-Stokes solver has been implemented using CUDA, adopting a collocated cartesian grid, SIMPLEC pressure-velocity coupling scheme, and implicit time discretization. The CUDA GPU implementations of the iterative solvers and preconditioners and the Navier-Stokes solver were validated and evaluated against serial and parallel CPU implementations. For the iterative solvers, speedups of between six and thirteen were achieved against the MKL CPU library, and the implemented methods beats existing open source GPU implementations of equivalent methods. For the full Navier-Stokes solver, speedups of up to a factor twelve were achieved compared to an equivalent commercial CPU code when equivalent iterative solvers were used. A speedup of a factor two was achieved when a commercial Algebraic MultiGrid method was used to solve the pressure Poisson equation in the commercial CPU implementation. The bottleneck of the resulting implementation was found to be the solution of the pressure Poisson equation. It accounted for a significant part of the total execution time for large problems. The implemented assembly routines on the GPU were highly efficient. The combined execution time for these routines were negligible compared to the total execution time. The GPU has been assessed as a highly capable accelerator for the implemented methods. About an order of magnitude of speedups have been achieved for algorithms which can efficiently be implemented on the GPU.
Animat Navigation Using Landmarks Navigation of Simulated Animals Inspired by Bees
(2018) Carlsson, Mathias; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Machine Learning techniques have made large advancements in previously challenging problems. A common problem with these methods is that they are very specialised on their specific field problem and it requires a lot of work to apply the methods on a new problem. A solution to this general issue is Artificial General Intelligence( AGI). This is an AI that is able to identify any problem and find a solution. Animals often needs to navigate complex environments to survive. This master thesis tries to implement a homing navigation model, used by bees to find food by remembering places relative positions to landmarks, in the general animat model. The model consists of using sensors that detect landmarks relative position to the animat. The model shows a small improvement when compared against Q-learning. The result suggests that the animat model can produce more sophisticated methods of navigation but further research needs to be conducted to explore its limits.
Anomaly Detection in PowerCells Auxiliary Power Unit
(2015) Hjortberg, Hampus; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
In the paper of Hayton et.al [1], One-class Support Vector Machine is used for health monitoring of a jet engine in order to discover when and if an abnormal event has occured. Hayton et.al used the amplitude of the vibration data from the engine shaft as the feature data to the One-class Support Vector Machine algorithm. This approach works well when the sensor data is known to be periodic, with a certain frequency; however it can not be used if the sensor data has an irregular shape. In this paper we will extend the concept of Hayton et.al [1] and use the Discrete Wavelet Transform coefficients as input data to the OCSVM, rather than the Fourier Transform. This way we are able to classify more arbitrary sensor data found in PowerCells Auxilliary Power Unit (APU). We will also introduce a novel approach of how to select the hyperparameter s for the Radial Basis Function Kernel, in order to avoid both overfitting and underfitting.
Breaking the Token Barrier: Leveraging Retrieval Methods to Facilitate Text-to-SQL on Extensive Tabular Data
(2024) Bergström, Victor; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Granath, Mats; Staron, Miroslaw
Technical analysis of systems is extensively utilized in the industrial sector, particularly in automotive development. This analysis is often conducted on logs generated during a system’s runtime, which typically contain sensor values, timestamps, and events. These logs represent the primary means for analyzing a system’s behavior and identifying failures. But, depending on their size and structure, finding the correct data in these tables can be difficult. At Zenseact, their large table with sensor data remains largely unused by developers due to the challenges associated with acquiring sufficient knowledge to leverage it effectively. Full utilization of this data could speed up development and allow for more robust software to be developed by enabling an easier way to find the cause of errors. This thesis suggests leveraging LLMs to perform text-to-SQL to interface with this table. Here, a developer could use natural language and the LLM would generate the SQL code and query the database. But current text-to-SQL methods are naive, they assume the table’s schemas are small enough to fit inside the LLM’s token limit which is not always the case, especially for Zenseact’s table with 5677 columns. This thesis tackles the overarching problem by crafting an artifact that employs retrieval methods to fetch only the related columns of the schema given a user’s natural language query. If this works, it also has the added benefit of simplifying the problem for the LLM where the columns it is presented with are fewer and relevant to the query. The performance is measured in Errorless Code ratio. This ratio measures how often the generated SQL queries can run, without errors, on the database. This does not take the correctness of the code into account. The results were measured using 0, 1, and 2 shot and came out to 35%, 34%, and 39% respectively. To improve performance, the problem was divided by letting the LLM in two separate steps, first select columns and then generate the SQL code. This approach improved the EC to 57%. The retrieval was benchmarked separately using 18 tests. Default retrieval, where the column is embedded and stored in a vector database yielded a result of 4 out of18 on the benchmark. To improve retrieval performance the columns were simplified using a term frequency removal algorithm. This improved the result to 6 out of 18 correct on the retrieval benchmark. The artifact presents reasonable approaches for text-to-SQL. Retrieval is a wellestablished method in research, and the 2-shot and refined column selection both improve a text-to-SQL system’s performance. However, using systems like this in real scenarios introduces a big hurdle for the models used. It’s almost impossible for the model to have learned enough in its initial training to understand the companyspecific information. The methods presented for text-to-SQL on extensive database tables could be viable, but further improvements to the AI models are required until they can be used in a production environment.
Cooperative Inverse Reinforcement Learning - Cooperation and learning in an asymmetric information setting with a suboptimal teacher
(2018) Ek, Johan; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
There exists many different scenarios where an artificial intelligence (AI) may have to learn from a human. One such scenario is when they both have to cooperate but only the human knows what the goal is. This is the study of cooperative inverse reinforcement learning (CIRL). The purpose of this report is to analyze CIRL when the human is not behaving fully optimally and may make mistakes. The effect of different behaviours by the human is investigated and two frameworks are developed, one for when there is a finite set of possible goals and one for the general case where the set of possible goals is infinite. Two benchmark problems are designed to compare the learning performance. The experiments show that the AI learns, but also that the humans behaviour has a large affect on learning. Also highlighted by the experiments, is the difficulty of differentiating between the actual goal and other possible goals that are similar in some aspects.
Data integration using machine learning: Automation of data mapping using machine learning techniques
(2016) Birgersson, Marcus; Hansson, Gustav; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Data integration involves the process of mapping the flow of data between systems. This is a task usually performed manually and much time can be saved if some parts of this can be automated. In this report three models based on statistics from earlier mapped systems is presented. The purpose of these models is to aid an expert in the mapping process by supplying a first guess on how to map two systems. The models are limited to mappings between two XML-formats, where the path to a node carrying data usually is descriptive of its data content. The developed models are the following: 1. A shortest distance model based on the concept that two nodes that have been associated with a third node but not each other most likely have something to do with each other. 2. A network flow model, which connects words with similar semantic meaning to be able to associate the words in two connected XML paths with each other. 3. A data value model which connects data values to nodes based on similarities between the value and earlier seen data. The results of the models agrees with expectations. The shortest distance model can only make suggestions based on XML-structures that are present in the training set supplied for the project. The network flow model has the advantage that it only needs to recognize parts of a path to map two nodes to each other, and even completely unfamiliar systems can be mapped if there are similarities between the two systems. Overall, the data value model performs the worst, but can make correct mappings in some cases when neither of the others can.
Discovering Novel Chemical Reactions
(2021) Rydholm, Emma; Svensson, Emma; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Damaschke, Peter; Haghir Chehreghani, Morteza
Accurately predicting chemical reactions can facilitate the search for optimal synthe sis routes in a chemical reaction network and as a consequence expedite the lengthy drug discovery process. As an effort in this direction, this work aims to explore AstraZeneca’s chemical knowledge graph by two complementary analyses. In a first part, graph theory related statistics is employed as a means to gain insights about the chemical reaction graph at AstraZeneca. Significant differences are observed be tween this internal reaction graph and the one based on the public dataset of United States patents as well as other reaction graphs discussed in literature. Secondly, a link prediction model is applied to and evaluated on AstraZeneca’s chemical reaction graph, in order to suggest new potential chemical reactions. In order to successfully accomplish this task, an existing link prediction model is adapted and trained. The test results are then compared to heuristic baselines, showing that the proposed implementation substantially exceeds what can be achieved with heuristic methods. One of the contribution from this research is a comparison between different ways to sample the ground truth class of non-existing links for training and evaluation. The choice of method for this task is shown to have an impact on the final predictions. Finally, a set of promising, predicted reactions are suggested and is currently under further investigation at AstraZeneca.
Distributed machine learning framework for shortest path problems with stochastic weights
(2023) Aspegrén, Gabriel; Nilsson Dahlberg, Olle; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Åkerblom, Niklas
Range anxiety is one of the most common reasons why customers hesitate to buy an electric car. At the same time, the European Union strives toward a fully electric car fleet. Previous work has shown promising results in designing a self-learning shortest path algorithm for finding the most energy efficient path for an electric vehicle through a road network, using combinatorial multi-armed bandit methods. However, it is desirable to scale the methods for larger networks, for example countries and continents. In this project, we design a distributed framework for shortest path computations on a road network, over a computer cluster, using combinatorial multi-armed bandit methods and machine learning. The system is distributed with Apache Spark and GraphX’s version of the Pregel algorithm. An experimental study is performed to investigate the impact of partition strategy, number of partitions, network size and latency between computer nodes on the total run-time. The results show that partitioning strategy has an significant impact on the run-time and that larger networks benefit more from being partitioned.
Dynamic State Representation for Homeostatic Agents
(2018) Mäkeläinen, Fredrik; Torén, Hampus; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
In a reinforcement learning setting an agent learns how to behave in an environment through interactions. For complex environments, the explorable state space can easily become unmanageable, and efficient approximations are needed. The Generic Animat model (GA model), heavily influenced by biology, takes an approach utilising a dynamic graph to represent the state. This thesis is part of the Generic Animat research project at Chalmers that develops the GA model. In this thesis, we identify and implement potential improvements to the GA model and make comparisons to standard Q-learning and deep Q-learning. With the improved GA model we show that in a state space larger than 232, we see substantial performance gains compared to the original model.
Estimating Causal Effects with Interpretable Decision Trees
(2023) Audinet De Pieuchon, Nicolas; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Damaschke, Peter; Johansson, Fredrik
In this work we explore three methods for estimating treatment effects from observational data using interpretable decision trees: the outcome variance tree, the propensity tree and the linear dependence tree. Each tree attempts to split the covariate space into balanced partitions from which treatment effects can be inferred. The outcome variance tree focuses on reducing the variance in the outcome variable, and makes use of a sensitivity analysis based on the residual standard deviation in the outcome. The propensity tree attempts to build a tree that approximates a separate estimate of the propensity score whilst remaining interpretable. The linear dependence tree measures the linear dependence in the partitions and attempts to minimize it directly. The three methods are compared, along with other benchmark methods, on two data sets: a synthetic data set generated from a simple model and the more realistic semi-synthetic IHDP data set. Performance is evaluated by comparing interval widths and coverage for confidence and sensitivity intervals. A functionally-grounded evaluation of interpretability is given with tree size as proxies. The results show that the outcome variance tree and the linear dependence tree perform better than the benchmarks in terms of sensitivity intervals but worse in terms of confidence intervals. The propensity tree however did not perform as well as expected and requires more work to better understand its behavior.
From Data to Descriptions: Efficient Data Retrieval in Autonomous Vehicle Development using Generative AI
(2024) Knapp, Jesper; Moberg, Klas; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Granath, Mats; Staron, Miroslaw
The software that enables autonomous driving requires a variety of sensors that generate a large amount of data. Data collected from a vehicle are often stored for later reference, either for testing new software components or to analyze the fleet of vehicles on a larger scale. Due to the large amount of varied data, finding a specific vehicle scenario in a collection of vehicle logs proves difficult. Current solutions mainly use SQL to query the database of logs, this solution does however require knowledge of both SQL and the specifics of the data that you are looking for. This thesis was carried out in collaboration with Zenseact, and aims to create an artifact called "Genius" to enable searching their logged vehicle data using natural language by applying generative AI to generate scenario descriptions. A scenario is a 30 second snippet of data from the vehicle logs that contain signals, which are the result of processed sensor data from the vehicles. Videos recorded from the roof of the vehicles complement the signal data. The descriptions are created in two parts, first images from the video are fed to LLaVA 1.5 7b, a multi-modal LLM that describes the scenario based on the image. A selection of key signals extracted from each log, as all signals cannot fit inside the context size of the deployed LLM, are then combined along with the image description and fed to a second LLM, Gemma 7b, to create a combined description. After the descriptions have been generated they are embedded using BGE-large, a text embedding model, and stored in ChromaDB to create a vector database. This database is then used to allow semantically searching the logs by comparing their distances in vector space to a natural language query. This study follows the design science research (DSR) methodology with three regulative cycles, with 5 phases in each, followed by a learnings section for each cycle with insights that are used in the subsequent cycle. Initial results with a smaller set of 8 scenarios show promising results in terms of how well the scenarios were separated in vector space, and the ability to search them using natural language. When scaling up to 100 scenarios, scenarios are mostly still searchable, however, scenarios that are not very distinct are hard to find since there are many similar matches. To counteract this, several systems of evaluating if the returned scenarios are correct were implemented, such as comparing keywords and an evaluation of the scenario distances. The generated descriptions were evaluated by engineers working with the vehicle logs at the collaborating company, on a scale of 1-5, the descriptions achieved a mean score of 3.3125. Overall, the solution can not replace existing solutions in its current form, this is due to the fact that all data is not available in the generated descriptions and the LLM and embedding models limited capabilities with numerical data.
Generative AI for Molecular Simulations
(2024) Chen, Weilong; Moqvist, Selma; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Olsson, Simon; Olsson, Simon
In statistical mechanics, computing the average behavior of microscopic states is crucial, for example, in estimating observables for equilibrium distributions in molecular systems. The challenge lies in the difficulty of sampling, as the density is known but hard to sample from. Typically, sampling of molecular conformations is performed using molecular dynamics, which faces challenges in obtaining iid samples due to the problem of rare events. Various enhanced sampling methods have been proposed to tackle this issue. Machine learning, specifically continuous-time generative models, offers a new perspective for tackling this problem. In our thesis, we propose two generative models using the recent Stochastic Interpolants framework. The first learns to transform between equilibrium distributions with different temperatures, which can be further applied with the current replica exchange method. The second model learns transition probability densities across time scales, which can be used as a surrogate model to accelerate MD simulations. We highlight the ability of Stochastic Interpolants to design efficient sampling methods for many-body systems in different ways, making it a powerful tool for advancing molecular simulation. Our results are two-fold. First, we present our Stochastic Interpolant ITO model and show how it reduces the VAMP-2 score gaps when benchmarked against the original ITO architecture. Next, we showcase our Thermodynamic Interpolant model, that to some extent manages to perform temperature transformations in a setting where it has to generalize beyond the training data. Our advancements show potential and could benefit various fields such as drug discovery, material science, catalysis, and green chemistry.
Graph Classification with Differential Privacy
(2015) Frost, Otto; Thufvesson Retzner, Carl; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
With increasing usage of online services such as email, social networks and online shopping more data than ever before is being gathered. Many types of data for which structure, flow and relationships are important may naturally be represented by graphs. Services may want to use machine learning algorithms to gain insight about their customers or users, and to successfully use the concepts of machine learning they must take the structure and properties of graph data into account. Seemingly innocuous data could potentially be used to infer sensitive information about individuals and should be kept private, the algorithms and techniques used must therefore take privacy into consideration. In this thesis we consider the combination of machine learning and privacy by bringing together the concepts of support vector machines and differential privacy. We examine the classification of graphs by means of kernel methods and present a framework for constructing private representations of two well known graph kernels, the random walk kernel and graphlet kernels. Furthermore, to allow for classification of large graphs we present a novel sampling scheme for approximation of subgraph counts. We evaluate both sampling and classification using four real world datasets consisting of social, road and protein networks.
Improved Pattern Generation for Bloom Filters with Bit Patterns Optimizing bit patterns for use in blocked Bloom filters
(2018) Hedström, Björn; Josefsson, Ivar; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
Set-membership is a commonly occurring problem in many areas of computing, from networking to database applications. A popular data structure for this problem is the Bloom filter: a small, hash-based probabilistic structure which guarantees no false negatives, but can result in false positives. Recently they have been used as an important tool in bioinformatics where the data sets are huge, and as a consequence the filters also need to be large. Blocked Bloom filters with bit patterns have been suggested as an alternative to cope with the deteriorated cache- and hash-behaviour in these cases. It was recently discovered that optimal pattern design for use in these structures is linked to two-stage group testing. There has also been some recent partial results that indicate a certain structure of optimal patterns. This thesis concerns itself with investigating these structural properties to find a better pattern design for use in Blocked Bloom filters with bit patterns. Our main result is a new, deterministic, algorithm for pattern generation used in these structures based on the Chinese Remainder Theorem. The results indicate that this construction improves the false positive rate for all our testing scenarios. As a side-result we also propose a modification to a known combinatorial design used in group testing which significantly reduces the needed number of tests for high number of defectives.
Latent Vector Synthesis
(2023) Högberg, David; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Björk, Staffan; Tatar, Kıvanç
Generative deep learning models for sound synthesis applications have gathered interest recently that are able to generate novel sound material based on the characteristics of a given audio dataset. A subcategory of these models are variational autoencoders, which build generative latent spaces of audio where sounds are organised based on similarity. Although expressive uses of these models abound, questions around their practical applicability and aesthetic implications as part of an artistic process remain underexplored. This thesis investigates the technological and aesthetic affordances of latent audio spaces in the context of creative sound design and exploration. To this end, a sound synthesis tool in the form of a latent vector synthesizer is conceptualised and developed from a first-person research through design perspective. The prototype addresses issues around real-time playability of current machine learning models for sound generation by training a variational autoencoder on short samples of audio signals. The generated waveforms are incorporated as part of a wavetable- and vector synthesis engine that enables timbral interpolations and explorations of sonic textures. Positioned at the intersection of sonic art and audio technology the design implementation uncovers some latent potentials and affordances of new technologies for musical tasks.
Learning to Play Games from Multiple Imperfect Teachers
(2014) Karlsson, John; Chalmers tekniska högskola / Institutionen för data- och informationsteknik (Chalmers); Chalmers University of Technology / Department of Computer Science and Engineering (Chalmers)
This project evaluates the modularity of a recent Bayesian Inverse Reinforcement Learning approach [1] by inferring the sub-goals correlated with winning board games from observations of a set of agents. A feature based architecture is proposed together with a method for generating the reward function space, making inference tractable in large state spaces and allowing for the combination with models that approximate stateaction values. Further, a policy prior is suggested that allows for least squares policy evaluation using sample trajectories. The model is evaluated on randomly generated environments and on Tic-tac-toe, showing that a combination of the intentions inferred from all agents can generate strategies that outperform the corresponding strategies from each individual agent.
Leveraging Large Language Models For System Log Analysis - Fault Troubleshooting Radio Units Using Log Data
(2024) Nir, Jacob; Snäll, William; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Damaschke, Peter; Haghir Chehreghani,, Morteza
This thesis investigates the application of large language models (LLMs) for system log analysis, specifically focusing on fault troubleshooting in radio units using log data. The primary objective is to enhance the efficiency and accuracy of system monitoring tools through state-of-the-art AI techniques. The research explores the utilization of retrieval-augmented generation (RAG) frameworks and parameterefficient fine-tuning (PEFT) methods to process and summarize log data. By employing pre-trained models such as Llama2, Llama3 and Mistral, the study evaluates different implementations to summarize segments of logs as well as extracting relevant information from them. The findings demonstrate that LLMs can significantly automate and improve the analysis of system logs, providing insights and facilitating easier troubleshooting. Additionally, the study examines the impact of enriching chatbot input data with contextual information, leading to substantial performance improvements in specialized domains. Despite the promising results, the research acknowledges limitations related to the quality and structure of log data and the need for source-specific refinements in context-enrichment methods. The contributions of this thesis are twofold: it presents a viable approach to leveraging LLMs for easier system monitoring and highlights the critical role of context in enhancing chatbot functionalities. Future research directions include integrating more advanced models, fine tuning existing models and exploring other state-of-theart methods to optimize retrieval-augmented generation pipelines.

Browse

Browsar Data- och informationsteknik (CSE) // Computer Science and Engineering (CSE) efter Program "Complex adaptive systems (MPCAS), MSc"

Sökresultat per sida

Sortera efter