Offered Theses

Partial response schemes like duobinary trade bandwidth for controlled interference between neighboring symbols, rather than eliminating it entirely. The student first works through the underlying theory and builds a simulation to verify it. Building on that, the student applies the concept to a realistic optical fiber link, where chromatic dispersion limits the usable bandwidth. Finally, the filter coefficients are optimized so that transmission reaches further than with the standard, fixed duobinary target.

Prerequisites

I regularly offer theses in the fields of probabilistic shaping, coding for high-throuput applications, and algorithms for communication. Please reach out if you are interested in a thesis in any of my research fields. A good background in information theory and channel coding are preferable, but the requirements generally depend on your interests.

Please include a description of your interests and corresponding academic background in your application. If you have a thesis idea, I am happy to discuss your suggestions. Also, I am available to supervise external theses as long as they are in my field of expertise.

Modern networked systems increasingly rely on intelligent information exchange between sensing devices and decision-making agents. Rather than sending data continuously or periodically, future communication networks aim to transmit only what matters: information that is useful, timely, and effective for achieving a specific goal.

Such goal-oriented communication is a key enabler of efficient cyber-physical systems, ranging from remote health monitoring and autonomous vehicles to industrial automation. Designing these systems requires new models that balance data freshness, communication cost, and decision accuracy.

This thesis will explore and model decision-making mechanisms for intelligent update systems where both sender and receiver actively decide when and what information to exchange. The goal is to investigate how coordinated or independent policies can improve overall system efficiency and effectiveness.

Possible directions include:

• Modeling joint decision processes between sensing and actuation agents.
• Analyzing when an agent should send (push) or request (pull) updates.
• Developing and simulating policies that account for usefulness, timeliness, and cost of communication.
• Comparing rule-based (model-based) and learning-based (reinforcement) approaches

Prerequisites

I may not always have prepared thesis topics available. Please feel free to reach out if you are interested in working on a thesis within any of my research areas.

Partial response schemes like duobinary trade bandwidth for controlled interference between neighboring symbols, rather than eliminating it entirely. The student first works through the underlying theory and builds a simulation to verify it. Building on that, the student applies the concept to a realistic optical fiber link, where chromatic dispersion limits the usable bandwidth. Finally, the filter coefficients are optimized so that transmission reaches further than with the standard, fixed duobinary target.

Prerequisites

Background

Homomorphic Encryption (HE) is a powerful cryptographic paradigm that allows computations to be performed directly on encrypted data, without ever needing to decrypt it first. The result of such a computation, once decrypted, is identical to what would have been obtained by performing the same operation on the plaintext. This makes HE particularly attractive for privacy-preserving applications in cloud computing, medical data analysis, and machine learning, where sensitive data must be processed by an untrusted third party.

Modern HE schemes — such as BGV, BFV, and CKKS — are built on the hardness of lattice problems, most notably Learning With Errors (LWE) and its ring variant (RLWE). Every ciphertext carries a small amount of noise, which is what makes the scheme hard to break but also what makes decryption imperfect and what grows with every homomorphic operation performed on the ciphertext. Managing this noise is the central practical bottleneck of HE.

The HE-Specific Gap

The research internship on this topic surveys how coding can close part of the gap between the uncoded message rate and the capacity of the LWE/MLWE channel — but only for a single, fresh ciphertext, exactly as in the PKE/KEM setting the two survey papers study. Fully Homomorphic Encryption is different in a way neither paper addresses: a scheme is standardly run as a leveled circuit of some fixed target multiplicative depth D, followed by bootstrapping, and the noise a packed message actually faces is not the noise of a single encryption but the accumulated result of D homomorphic operations. Moreover, whatever encoding is used to pack the message must remain meaningful throughout — homomorphic addition and multiplication act algebraically on the encoded plaintext directly, so an error-correcting code layered on top of the message has to commute with those operations, or be explicitly designed around what they do to it. Neither constraint has a counterpart in the PKE/KEM literature, where the ciphertext is decoded immediately after a single noisy hop.

Thesis Task

Building on the internship's survey, the student will investigate the depth-D LWE/ channel under the additional constraint that the message encoding must commute with homomorphic addition and multiplication. Rather than aiming directly for an explicit optimal code — which the homomorphism-compatibility constraint makes considerably harder to construct than the codes surveyed in the internship — the student will approach the problem from a bounds perspective:

• characterize the channel faced by a packed message at a fixed target depth D, extending the existing capacity results for the fresh-ciphertext LWE channel to this accumulated-noise setting;
• formalize the class of homomorphism-compatible encoders (e.g., module or ring homomorphisms into the plaintext space) and derive achievability and converse bounds on the best rate attainable within this constrained class, in the spirit of structured-coding results such as Körner–Marton and compute-and-forward;
• quantify the resulting gap between this constrained capacity and the unconstrained channel capacity, to establish how much rate is fundamentally given up for homomorphism-compatibility alone;
• where possible, check whether simple existing constructions (e.g., linear or CRT/SIMD-based plaintext packing already used in BGV/BFV) approach the derived bound

References

[1] Regev, O. "On Lattices, Learning with Errors, Random Linear Codes, and Cryptography." STOC, 2005.

[2] Maringer, G. and Wachter-Zeh, A. "Reducing Ciphertext and Key Sizes for MLWE-Based Cryptosystems." arXiv:2502.01339, 2025.

[3] Lee, E., Kim, Y.-S., No, J.-S., Song, M., and Shin, D.-J. "Modification of FrodoKEM Using Gray and Error-Correcting Codes." IEEE Access, vol. 7, pp. 179564–179574, 2019.

[4] Maringer, G., Puchinger, S., and Wachter-Zeh, A. "Information- and Coding-Theoretic Analysis of the RLWE/MLWE Channel." IEEE Transactions on Information Forensics and Security, vol. 18, pp. 549–564, 2022.

[5] Gentry, C. "Fully Homomorphic Encryption Using Ideal Lattices." STOC, pp. 169–178, 2009.

[6] Brakerski, Z., Gentry, C., and Vaikuntanathan, V. "(Leveled) Fully Homomorphic Encryption without Bootstrapping." ITCS, pp. 309–325, 2012.

[7] Körner, J. and Marton, K. "How to Encode the Modulo-Two Sum of Binary Sources." IEEE Transactions on Information Theory, vol. 25, no. 2, pp. 219–221, 1979.

[8] Nazer, B. and Gastpar, M. "Compute-and-Forward: Harnessing Interference Through Structured Codes." IEEE Transactions on Information Theory, vol. 57, no. 10, pp. 6463–6486, 2011.

[9] Lyu, S., Liu, L., Ling, C., Lai, J., and Chen, H. "Lattice Codes for Lattice-Based PKE." Designs, Codes and Cryptography, 2023.

Prerequisites

See more information and apply directly here: Intern/Master Thesis Student - Topic/Area - Huawei Research Center Germany

Background

Homomorphic Encryption (HE) is a powerful cryptographic paradigm that allows computations to be performed directly on encrypted data, without ever needing to decrypt it first. The result of such a computation, once decrypted, is identical to what would have been obtained by performing the same operation on the plaintext. This makes HE particularly attractive for privacy-preserving applications in cloud computing, medical data analysis, and machine learning, where sensitive data must be processed by an untrusted third party.

The Noise Problem

Modern HE schemes — such as BGV, BFV, and CKKS — are built on the hardness of lattice problems, most notably Learning With Errors (LWE) and its ring variant (RLWE). The security of these schemes relies on the presence of noise in ciphertexts. However, homomorphic operations cause this noise to grow. Left unchecked, this noise growth quickly renders ciphertexts undecryptable, fundamentally limiting the depth and complexity of computations that can be performed. Noise management is therefore a central challenge in practical HE.

Internship Task

Several techniques have been developed to control or reduce noise growth in HE schemes, such as bootstrapping, modulus switching, flattening and rescaling.

In this theis, the student will try to develop a new noise management technique. The student will then compare this new technique to the existing ones, compare their computational costs and performance. The student will complement this with practical experiments using established HE libraries.

References

[1] Gentry, C. "Fully Homomorphic Encryption Using Ideal Lattices." Proceedings of the 41st Annual ACM Symposium on Theory of Computing (STOC), pp. 169–178, 2009.

[2] Brakerski, Z. and Vaikuntanathan, V. "Fully Homomorphic Encryption from Ring-LWE and Security for Key Dependent Messages." Advances in Cryptology – CRYPTO 2011, Lecture Notes in Computer Science, vol. 6841, Springer, 2011.

[3] Brakerski, Z., Gentry, C., and Vaikuntanathan, V. "(Leveled) Fully Homomorphic Encryption without Bootstrapping." Proceedings of the 3rd Innovations in Theoretical Computer Science Conference (ITCS), pp. 309–325, 2012.

[4] Fan, J. and Vercauteren, F. "Somewhat Practical Fully Homomorphic Encryption." Cryptology ePrint Archive, Report 2012/144, 2012.

[5] Cheon, J. H., Kim, A., Kim, M., and Song, Y. "Homomorphic Encryption for Arithmetic of Approximate Numbers." Advances in Cryptology – ASIACRYPT 2017, Lecture Notes in Computer Science, vol. 10624, Springer, pp. 409–437, 2017.

[6] Gentry, C., Sahai, A., and Waters, B. "Homomorphic Encryption from Learning with Errors: Conceptually Simpler, Asymptotically Faster, Attribute-Based." Advances in Cryptology – CRYPTO 2013, Lecture Notes in Computer Science, vol. 8042, Springer, pp. 75–92, 2013.

Post-quantum cryptography (PQC) has been an active area of research since the seminal work of Shor. 
Indeed, most public-key cryptography currently deployed (such as RSA and elliptic-curve-based schemes) is vulnerable to quantum adversaries. 
This applies in particular to public-key encryption (PKE) schemes.  
An attacker could record encrypted traffic and later decrypt it once a sufficiently capable quantum computer becomes available - a strategy known as "harvest now, decrypt later."

Post-quantum secure alternatives can be constructed from hard problems on codes and lattices. The recently standardized Kyber and HQC follow an encryption-based approach, while, e.g., NewHope is based on a key reconciliation mechanism:
Alice and Bob obtain noise variants of a common secret, and error correction removes this noise, allowing them to agree on the same shared key.

This project will survey constructions for exchanging a key in code-based and lattice-based cryptography. These constructions are to be categorized based on key properties, such as underlying metric, bandwidth requirements, and underlying assumptions.
First, an overview of the key techniques is to be developed.
Then, you are going to analyze whether they can be transferred from codes to lattices and vice versa. The project requires reading and understanding the literature on lattice- and code-based KEMs; a good starting point can be the following works:

Aguilar-Melchor, Carlos, et al. "Efficient encryption from random quasi-cyclic codes." IEEE Transactions on Information Theory 64.5 (2018): 3927-3943.

Bos, Joppe, et al. "CRYSTALS-Kyber: a CCA-secure module-lattice-based KEM." 2018 IEEE European symposium on security and privacy (EuroS&P). IEEE, 2018.

Alkim, Erdem, et al. "Post-quantum Key Exchange — A new hope." 25th USENIX security symposium (USENIX Security 16). 2016.

Prerequisites

I regularly offer theses in the fields of probabilistic shaping, coding for high-throuput applications, and algorithms for communication. Please reach out if you are interested in a thesis in any of my research fields. A good background in information theory and channel coding are preferable, but the requirements generally depend on your interests.

Please include a description of your interests and corresponding academic background in your application. If you have a thesis idea, I am happy to discuss your suggestions. Also, I am available to supervise external theses as long as they are in my field of expertise.

Wireless communication links used in control and automation systems must deliver timely and reliable information despite fading and packet losses. Hybrid Automatic Repeat reQuest (HARQ) protocols -combining forward error correction and retransmissions- can significantly improve reliability. However, retransmitting outdated packets can delay fresh information, leading to performance degradation in closed-loop systems.

This thesis aims to analyze and optimize HARQ scheduling for networked control from a communication-theoretic perspective. The project will study how retransmission strategies, feedback mechanisms, and channel dynamics jointly influence information freshness and system performance. The problem will be formulated within a stochastic decision framework, such as a Markov decision process (MDP), to characterize the trade-off between reliability, latency, and communication cost.

Possible research directions include:

Designing scheduling strategies that decide between new transmissions and retransmissions; 



The thesis combines ideas from communication theory, stochastic modeling, and decision-making under uncertainty, and provides an opportunity to contribute to ongoing research in low-latency and reliable communication for cyber-physical systems.

Prerequisites

Federated learning allows to train a machine learning model in a distributed manner, i.e., the training data are collected and stored locally by users such as mobile devices or multiple institutes. The training is under the coordination of a central server and performed iteratively. In each iteration, the server sends the current global model to the users, who update their local model and send the local updates to the server for aggregation.

FL is proposed to protect user's sensitive data since these training data never leave the user devices. However, works have shown that the local updates still leaks information about the local datasets. To deal with this leakage, SecAgg[1] is proposed. Secure aggregation is to make sure that the server only obtains the aggregation of the local updates rather than each individual update.

However, recent work [2] has shown that, SecAgg only preserves privacy of the users in a single training round. Due to user selection in federated learning, by observing the aggregated models over multiple training rounds, the server is able to recoverindividual local models of the users.

The goal of this seminar is to study and understand SecAgg [1], the multi-round privacy leakage it suffers and how is this problem solved in [2]. 

[1]. Bonawitz, Keith, et al. "Practical secure aggregation for privacy-preserving machine learning." proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security. 2017.

[2]. So, Jinhyun, et al. "Securing secure aggregation: Mitigating multi-round privacy leakage in federated learning." Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. No. 8. 2023.

How many ways are there to write down n nonnegative integers, all of which being strictly smaller than q, such that their sum is k? In other words, what is the coefficient of x^k in (1 + x + ... + x^(q-1))^n?

In this thesis, we are going to dive into the integer partitioning problem with a certain number of partitions and an upper bound on partition size.

The goal of this thesis is to take an already existing bound for the value mentioned above, which holds for a specific k value - and extend it to general k. 

The upper bound can then be used for proving better upper bounds in coding theory for certain metrics other than the Hamming metric.

[1] H. B.-S. Couvée, T. Jerkovits, and J. Bariffi, ‘Bounds on Sphere Sizes in the Sum-Rank Metric and Coordinate-Additive Metrics’, Des. Codes Cryptogr., Mar. 2025, doi: 10.1007/s10623-025-01604-0.

Prerequisites

I may not always have prepared thesis topics available. Please feel free to reach out if you are interested in working on a thesis within any of my research areas.

Motivated by the cell-free and massive MIMO (multiple input multiple outputs) communication scenarios, the number of power amplifiers (PA), digital to analog converters (DAC), etc., is increased. Thus, using coarse quantized transmission reduces the channel's hardware cost and nonlinear effects. More details can be found here.

In this project, we investigate algorithms for mapping modulated data to coarsely quantized signals. We potentially employ linear or non-linear precoding techniques and receiver architectures, comparing our results with classical methods and neural networks.

The student needs an understanding of information theory and communication systems.

Partial response schemes like duobinary trade bandwidth for controlled interference between neighboring symbols, rather than eliminating it entirely. The student first works through the underlying theory and builds a simulation to verify it. Building on that, the student applies the concept to a realistic optical fiber link, where chromatic dispersion limits the usable bandwidth. Finally, the filter coefficients are optimized so that transmission reaches further than with the standard, fixed duobinary target.

Prerequisites

Background

Homomorphic Encryption (HE) is a powerful cryptographic paradigm that allows computations to be performed directly on encrypted data, without ever needing to decrypt it first. The result of such a computation, once decrypted, is identical to what would have been obtained by performing the same operation on the plaintext. This makes HE particularly attractive for privacy-preserving applications in cloud computing, medical data analysis, and machine learning, where sensitive data must be processed by an untrusted third party.

The Noise Problem

Modern HE schemes — such as BGV, BFV, and CKKS — are built on the hardness of lattice problems, most notably Learning With Errors (LWE) and its ring variant (RLWE). The security of these schemes relies on the presence of noise in ciphertexts. However, homomorphic operations cause this noise to grow. Left unchecked, this noise growth quickly renders ciphertexts undecryptable, fundamentally limiting the depth and complexity of computations that can be performed. Noise management is therefore a central challenge in practical HE.

Internship Task

Several techniques have been developed to control or reduce noise growth in HE schemes, such as bootstrapping, modulus switching, flattening and rescaling.

In this internship, the student will conduct a structured investigation of noise management techniques in homomorphic encryption. The goal is to develop a thorough and hands-on understanding of the state of the art. Concretely, the student will survey the techniques described above, study their theoretical underpinnings, compare their computational costs and noise behavior across different HE schemes (e.g., BGV, BFV, CKKS), and ideally complement this with practical experiments using an established HE library such as Microsoft SEAL or OpenFHE.

References

[1] Gentry, C. "Fully Homomorphic Encryption Using Ideal Lattices." Proceedings of the 41st Annual ACM Symposium on Theory of Computing (STOC), pp. 169–178, 2009.

[2] Brakerski, Z. and Vaikuntanathan, V. "Fully Homomorphic Encryption from Ring-LWE and Security for Key Dependent Messages." Advances in Cryptology – CRYPTO 2011, Lecture Notes in Computer Science, vol. 6841, Springer, 2011.

[3] Brakerski, Z., Gentry, C., and Vaikuntanathan, V. "(Leveled) Fully Homomorphic Encryption without Bootstrapping." Proceedings of the 3rd Innovations in Theoretical Computer Science Conference (ITCS), pp. 309–325, 2012.

[4] Fan, J. and Vercauteren, F. "Somewhat Practical Fully Homomorphic Encryption." Cryptology ePrint Archive, Report 2012/144, 2012.

[5] Cheon, J. H., Kim, A., Kim, M., and Song, Y. "Homomorphic Encryption for Arithmetic of Approximate Numbers." Advances in Cryptology – ASIACRYPT 2017, Lecture Notes in Computer Science, vol. 10624, Springer, pp. 409–437, 2017.

[6] Gentry, C., Sahai, A., and Waters, B. "Homomorphic Encryption from Learning with Errors: Conceptually Simpler, Asymptotically Faster, Attribute-Based." Advances in Cryptology – CRYPTO 2013, Lecture Notes in Computer Science, vol. 8042, Springer, pp. 75–92, 2013.

Machine-learning (ML) algorithms become increasingly popular tools for various ap-
plications. Since the hardware requirements to train ML algorithms is heavy, we of-
ten rely on service providers that provide pre-trained ML inference interfaces. To
ensure security of input data, we would like to send the our data in encrypted form
so that neither during transmission, nor the service provider can learn meaningful
information from the transmission and computations, but keeping up the function-
alities provided by ML algorithms.

Fully homomorphic encryption (FHE), an encryption scheme that allows various
computations on encrypted data, enables secure ML inference. Indeed, the user
who wishes to use a ML inference service can merely encrypt their data using an
FHE protocol. The service provider that knows, which FHE protocol has been used,
can run the encrypted version of the ML algorithm on the encrypted data and sends
back the encrypted response. Finally, the user decrypts and receives the inference
result.

In this project, we want to focus on one of the main parts of the ML inference steps,
namely, matrix-vector multiplication. Here, the matrix is a result of the training and
stored in plaintext on the server side. The vector is the user’s input data. During this
research internship, we assess the performance of different FHE protocols based on
lattices for the task of matrix-vector multiplication. Depending on the initial results,
we try to increase the efficiency using error correcting codes in lattice-based FHE
schemes, or to tailor the schemes to the present arithmetic task of matrix-vector
multiplication to increase the performance.

Secure Training for Machine-Learning
Note that this research internship only asks for a secure inference step. Training on
encrypted data, on the other hand, will not be feasible using FHE algorithms. In-
deed, while FHEs allow for general encrypted computations, the overhead increases
with the depth of the arithmetic circuits to be computed. The complexity of training
algorithms and the sheer amount of data required for training, makes encrypted
computations unrealistic in practice.

The MinRank and Rank Syndrome Decoding problem are the basis of the security of many cryptographic schemes. Therefore it is important to understand the complexity of solving these problems.

In this project, the goal is to:
1. Understand the the existing solvers for these problems, in particular, [1] and [2].
2. Apply some ideas to slightly improve their complexity.
3. Understand if it affects the security of the existing cryptographic schemes.

[1] Gaborit, Philippe, Olivier Ruatta, and Julien Schrek. 2016. “On the Complexity of the Rank Syndrome Decoding Problem.” IEEE Transactions on Information Theory 62 (2): 1006–19. https://doi.org/10.1109/TIT.2015.2511786.
[2] L. Goubin and N. T. Courtois, “Cryptanalysis of the TTM Cryptosystem,” in Advances in Cryptology — ASIACRYPT 2000, T. Okamoto, Ed., Berlin, Heidelberg: Springer, 2000, pp. 44–57. https://doi.org/10.1007/3-540-44448-3_4.

Prerequisites

Post-quantum cryptography (PQC) has been an active area of research since the seminal work of Shor. 
Indeed, most public-key cryptography currently deployed (such as RSA and elliptic-curve-based schemes) is vulnerable to quantum adversaries. 
This applies in particular to public-key encryption (PKE) schemes.  
An attacker could record encrypted traffic and later decrypt it once a sufficiently capable quantum computer becomes available - a strategy known as "harvest now, decrypt later."

Post-quantum secure alternatives can be constructed from hard problems on codes and lattices. The recently standardized Kyber and HQC follow an encryption-based approach, while, e.g., NewHope is based on a key reconciliation mechanism:
Alice and Bob obtain noise variants of a common secret, and error correction removes this noise, allowing them to agree on the same shared key.

This project will survey constructions for exchanging a key in code-based and lattice-based cryptography. These constructions are to be categorized based on key properties, such as underlying metric, bandwidth requirements, and underlying assumptions.
First, an overview of the key techniques is to be developed.
Then, you are going to analyze whether they can be transferred from codes to lattices and vice versa. The project requires reading and understanding the literature on lattice- and code-based KEMs; a good starting point can be the following works:

Aguilar-Melchor, Carlos, et al. "Efficient encryption from random quasi-cyclic codes." IEEE Transactions on Information Theory 64.5 (2018): 3927-3943.

Bos, Joppe, et al. "CRYSTALS-Kyber: a CCA-secure module-lattice-based KEM." 2018 IEEE European symposium on security and privacy (EuroS&P). IEEE, 2018.

Alkim, Erdem, et al. "Post-quantum Key Exchange — A new hope." 25th USENIX security symposium (USENIX Security 16). 2016.

Prerequisites

I regularly offer theses in the fields of probabilistic shaping, coding for high-throuput applications, and algorithms for communication. Please reach out if you are interested in a thesis in any of my research fields. A good background in information theory and channel coding are preferable, but the requirements generally depend on your interests.

Please include a description of your interests and corresponding academic background in your application. If you have a thesis idea, I am happy to discuss your suggestions. Also, I am available to supervise external theses as long as they are in my field of expertise.

I may not always have prepared thesis topics available. Please feel free to reach out if you are interested in working on a thesis within any of my research areas.

Motivated by the cell-free and massive MIMO (multiple input multiple outputs) communication scenarios, the number of power amplifiers (PA), digital to analog converters (DAC), etc., is increased. Thus, using coarse quantized transmission reduces the channel's hardware cost and nonlinear effects. More details can be found here.

In this project, we investigate algorithms for mapping modulated data to coarsely quantized signals. We potentially employ linear or non-linear precoding techniques and receiver architectures, comparing our results with classical methods and neural networks.

The student needs an understanding of information theory and communication systems.

I regularly offer theses in the fields of probabilistic shaping, coding for high-throuput applications, and algorithms for communication. Please reach out if you are interested in a thesis in any of my research fields. A good background in information theory and channel coding are preferable, but the requirements generally depend on your interests.

Please include a description of your interests and corresponding academic background in your application. If you have a thesis idea, I am happy to discuss your suggestions. Also, I am available to supervise external theses as long as they are in my field of expertise.

I may not always have prepared thesis topics available. Please feel free to reach out if you are interested in working on a thesis within any of my research areas.

Motivated by the cell-free and massive MIMO (multiple input multiple outputs) communication scenarios, the number of power amplifiers (PA), digital to analog converters (DAC), etc., is increased. Thus, using coarse quantized transmission reduces the channel's hardware cost and nonlinear effects. More details can be found here.

In this project, we investigate algorithms for mapping modulated data to coarsely quantized signals. We potentially employ linear or non-linear precoding techniques and receiver architectures, comparing our results with classical methods and neural networks.

The student needs an understanding of information theory and communication systems.