Efficient & Effective AI Systems
Bayesian active learning, cost-aware data acquisition, uncertainty quantification, adaptive evaluation, and principled model selection under resource constraints.
I develop efficient and effective AI systems under resource constraints, with a focus on Bayesian active learning, cost-aware learning, uncertainty quantification, adaptive evaluation, and applications in NLP, healthcare, and scientific discovery.
Bayesian Active LearningCost-aware LearningUncertainty QuantificationAdaptive EvaluationNLP & LLMsHealthcare AIAI for Science
Research
My work studies how AI systems can learn, adapt, and be evaluated effectively when data, labels, compute, or other resources carry real costs.
NLP, LLMs & Multilingual AI
Language modeling, speech, machine translation, retrieval, question answering, personalization, memory, and evaluation in multilingual and low-resource settings.
Healthcare & AI for Science
Maternal health, medical imaging, biological sequence design, genomic modeling, drug discovery, and scientific problems where reliability and acquisition costs matter.
Language technology impactBuilt foundational Fon resources and systems, beginning with FFR and followed by work on tokenization, embeddings, speech recognition, and multitask learning. Together, these contributions helped build the technical and data foundation that led to Fon being integrated into Google Translate in 2024.
Research across the stackWork spans foundational methods, LLMs, healthcare, speech, 3D medical imaging, generative models, biological sequence design, genomic modeling, and production-oriented AI systems.
Selected Projects & Technical Contributions
Key contribution
FFR and the Fon language technology stack
Created FFR, the foundational Fon-French neural machine translation system and corpus. Subsequent work on Fon tokenization, AfriVec representations, OkwuGbe speech recognition, and FonMTL helped build the research and resource ecosystem that led to Fon being integrated into Google Translate in 2024.
Fon Automatic Speech Recognition
Developed end-to-end automatic speech recognition for Fon as part of OkwuGbe, extending speech technology to a language with very limited labeled audio resources.
MMTAfrica
Co-created the first multilingual machine translation system in this line of work for six African languages, advancing cross-lingual transfer in low-resource settings.
OkwuGbe Python ASR Library
Built and released a Python library that makes low-resource automatic speech recognition models easier to use and integrate in downstream applications.
AfriVec
Created word embedding models for African languages, including Fon and Nobiin, to improve reusable lexical representations for downstream NLP.
DeepSNPs
Developed deep learning models for classifying disease-associated single nucleotide polymorphisms from chromosome locations, connecting representation learning with bioinformatics.
AfroLM
Developed an Afrocentric multilingual language model for 23 African languages using self-active learning, designed for transfer across multiple downstream NLP tasks.
Education
McGill University
PhD Candidate, Computer Science
2023 - 2026 | Montreal, Canada | Supervisor: Prof. Jackie Cheung
PhD thesis: Active Learning and Evaluation in Resource-Constrained Settings: Applications to African Languages.
Graduate coursework
- Representation Learning: A+ (Winter 2023)
- Causality and Machine Learning: A- (Winter 2023)
- Statistical Methods for Computer Vision: A- (Winter 2023)
- Applied Machine Learning: B+ (Winter 2024)
Jacobs University Bremen
MSc with Honors, Computer Science and Data Engineering
2020 - 2022 | Germany | 130 ECTS | GPA 4/4 Canadian equivalent
Thesis: DeepSNPs: Deep Learning for Disease SNPs Classification based on Chromosome Locations.
Recognition: Dean's Prize for Outstanding Master's Thesis.
Kazan Federal University
BSc with Honors, Mathematics
2016 - 2020 | Kazan, Russia | GPA 3.9/4 Canadian equivalent
- Graduated in the top 1% of class.
- Dean's List 2016-2020 and State Academic Scholarship.
- Best Bachelor Thesis: Speech Emotion Recognition with Deep Convolutional Networks.
- Elected Best Mathematics Student and Best International Student, 2016-2020.
Research & Work Experience
Complete research and professional experience from my CV, with the technical scope of each role retained.
Research Scientist Intern / Applied Scientist II
Amazon Research (Alexa+)
- Conducted research on a temporal self-evolving memory graph to help Alexa+ understand and retain user context over time.
- Leveraged a reinforcement learning agent to dynamically manage the addition, update, and deprecation of memory nodes.
- Worked on advanced personalization for LLM-powered conversational AI systems.
STC AI Scientist
The World Bank Group
- Prototype and refine research-grade AI/ML models for applied development contexts.
- Design, implement, and evaluate model architectures across multiple methodological variants.
- Conduct systematic benchmarking and comparative analysis of model performance.
- Fine-tune large language models for domain-specific applications.
- Develop rigorous evaluation frameworks, including diagnostic assessments, out-of-sample generalization tests, and scalability analyses.
- Collaborate with principal investigators and subject-matter experts to ensure methodological rigor and real-world relevance.
- Work with AI/ML engineers to translate research models into efficient, low-latency production systems.
- Synthesize, interpret, and communicate empirical findings to technical and non-technical stakeholders.
- Contribute to academic publications and research dissemination.
- Integrate emerging advances in AI/ML into ongoing work.
Machine Learning Research Intern
Google Research
- Conducted research on large language models for critical healthcare challenges, with a focus on women's and maternal health.
- Developed NLP models to predict postpartum mood disorders and other maternal health outcomes.
- Designed robust and efficient solutions aimed at future integration into clinical workflows.
- Work resulted in a paper accepted at Machine Learning for Healthcare 2025 and under preparation for Nature Medicine.
- Work led to a Google patent application.
Fundamental AI Research Scientist
Lelapa AI
- Led research on NLP methods tailored to African languages and their linguistic and cultural nuances.
- Designed and optimized models for language modeling, sentiment analysis, and machine translation in resource-constrained settings.
- Developed speech recognition systems for low-resource languages such as Sesotho that outperformed state-of-the-art benchmarks.
- Collaborated with data scientists and linguists to analyze African linguistic data and build multilingual evaluation frameworks.
- Contributed to algorithmic innovation, prototyping, production integration, technical documentation, and publications.
Machine Learning Research Intern
RBC Borealis AI
- Researched methods to improve the internal consistency and reliability of large language models across extended interactions.
- Explored techniques for improving robustness and logical coherence beyond surface-level accuracy.
- Investigated approaches toward safer and more dependable real-world LLM deployment.
- Focused on translating cutting-edge research into practical and trustworthy NLP systems.
AI Scientist in Residence
Probe Medical
- Applied self-supervised learning to disease detection and clinical report generation for 3D CT medical volumes such as CT-RATE.
- Designed models using Masked Autoencoders and Contrastive Predictive Coding to improve accuracy and diagnostic relevance.
- Worked toward context-aware medical report generation directly from imaging data.
Graduate Student Researcher (Volunteer)
Masakhane Research Foundation
- Conducted research on scalable NLP systems spanning machine translation, speech recognition, information retrieval, question answering, and large language models for low-resource African languages.
- Mentored undergraduate students and new community members, supporting first research publications and participation in the African NLP community.
- Contributed to Masakhane's mission of inclusive and collaborative African NLP research at scale.
Machine Learning Research Scientist Consultant
Phagos Biotech
- Built large-scale language models for genomic sequencing and small-molecule generation.
- Applied LLMs to learn phage structure and derive robust, invariant representations.
- Used learned representations for downstream tasks such as predicting toxicity levels from phage-bacteria interactions.
NLP Student Researcher
Google Research
- Built dataset infrastructure and processing pipelines for Named Entity Recognition in 20 African languages.
- Explored contextualized vocabularies for more than 200 African languages to improve representations and downstream task performance.
- Pretrained and adapted mT5 models on large-scale multilingual corpora and evaluated them on NER and grapheme-to-phoneme tasks.
Deep Learning and Drug Discovery Researcher
Mila - Quebec AI Institute
- Conducted foundational machine learning research for drug discovery under Prof. Yoshua Bengio and Dr. Dianbo Liu.
- Developed an active learning framework and graph-based selection method to filter and generate promising protein sequences.
- Worked on biological sequence design using Generative Flow Networks, contributing to an ICML 2022 publication.
- Contributed to GFlowNet-based smart dropout methods accepted at ICML 2023.
NLP Data Scientist
Roche Canada
- Led projects that delivered production-level code, improving runtime efficiency and team deliverables by a factor of 10.
- Built multilingual translation models enabling patients to report medication feedback in their native languages.
- Developed deep learning models to detect and classify anomalies in pharmaceutical production processes.
Part-time Data Scientist
Speeqo
- Crawled and preprocessed speech emotion datasets including EmoVo, Emodb, Savee, and Ravdess.
- Built speech denoising and enhancement models to improve signal robustness.
- Worked on speech synthesis and analysis to detect stress in children's recordings and study its relationship with academic performance.
Scientist in Residence
Modelis (James Doyle)
- Crawled, cleaned, and processed chemical compound databases for model training.
- Developed a generative model using SMILES representations to produce valid, novel, and diverse compounds within an active learning framework.
- Designed a Graph Convolutional Network to predict compound control percentages and z-scores.
First-author & Co-first-author Publications
Homepage publication list restricted to papers where I am first author or explicitly co-first author. The complete publication record remains on Google Scholar and Semantic Scholar.
COLM 2026
RankBALD: Ranking-Aligned Active Evaluation for Language Models
Bonaventure F. P. Dossou, Jackie Cheung
Bonaventure F. P. Dossou, Jackie Cheung
EACL 2026
Active Learning with Non-Uniform Costs for African Natural Language Processing
Bonaventure F. P. Dossou, Ines Arous, Audrey Durand, Jackie Cheung
Bonaventure F. P. Dossou, Ines Arous, Audrey Durand, Jackie Cheung
EMNLP 2025
MLHC 2025
Early Prediction of Postpartum Mood Disorders from Longitudinal Wearable Biometrics using deep learning and time series generative adversarial networks
Bonaventure F. P. Dossou, Mercy Nyamewaa Asiedu, Maja Mataric, Katherine A. Heller, Belen Lafon, Nichole Young-Lin
Bonaventure F. P. Dossou, Mercy Nyamewaa Asiedu, Maja Mataric, Katherine A. Heller, Belen Lafon, Nichole Young-Lin
ACL 2025
ACL 2025
Rethinking Full Finetuning from Pretraining Checkpoints in Active Learning for African Languages
Bonaventure F. P. Dossou, Ines Arous, Jackie Cheung
Bonaventure F. P. Dossou, Ines Arous, Jackie Cheung
Preprint / under review
InkubaLM: A small language model for low-resource African languages
Atnafu Lambebo Tonja*, Bonaventure F. P. Dossou*, Jessica Ojo, Jenalea Rajab, Fadel Thior, Eric Peter Wairagala, Aremu Anuoluwapo, Pelonomi Moiloa, Jade Abbott, Vukosi Marivate, Benjamin Rosman (*equal contribution)
Atnafu Lambebo Tonja*, Bonaventure F. P. Dossou*, Jessica Ojo, Jenalea Rajab, Fadel Thior, Eric Peter Wairagala, Aremu Anuoluwapo, Pelonomi Moiloa, Jade Abbott, Vukosi Marivate, Benjamin Rosman (*equal contribution)
EMNLP 2023
FonMTL: Towards Multitask Learning for the Fon Language
Bonaventure F. P. Dossou, Iffanice Houndayi, Pamely Zantou, Gilles HACHEME
Bonaventure F. P. Dossou, Iffanice Houndayi, Pamely Zantou, Gilles HACHEME
ICLR 2023
Pretrained Vision Models for Predicting High-Risk Breast Cancer Stage
Bonaventure F. P. Dossou, Yeno Gbenou, Miglanche Ghomsi
Bonaventure F. P. Dossou, Yeno Gbenou, Miglanche Ghomsi
EMNLP 2022
Self-Active Learning for Multilingual Language Models: Case Study of 23 African Languages
Bonaventure F. P. Dossou, Atnafu Lambebo Tonja, Oreen Yousuf, Salomey Osei, Abigail Oppong, Iyanuoluwa Adeola Shode, Oluwabusayo Olufunke Awoyomi
Bonaventure F. P. Dossou, Atnafu Lambebo Tonja, Oreen Yousuf, Salomey Osei, Abigail Oppong, Iyanuoluwa Adeola Shode, Oluwabusayo Olufunke Awoyomi
Preprint
GraphCC for Diverse and Novel Antimicrobial Peptides Generation and Selection
Bonaventure F. P. Dossou, Dianbo Liu, Xu Ji, Moksh Jain, Almer M. van der Sloot, Roger Palou, Michael Tyers, Yoshua Bengio
Bonaventure F. P. Dossou, Dianbo Liu, Xu Ji, Moksh Jain, Almer M. van der Sloot, Roger Palou, Michael Tyers, Yoshua Bengio
ICCV 2021
FSER: Deep Convolutional Neural Networks for Speech Emotion Recognition
Bonaventure F. P. Dossou, Yeno Gbenou
Bonaventure F. P. Dossou, Yeno Gbenou
EMNLP 2021
EACL 2021
AfriVec: Word Embedding Models for African Languages. Case Study of Fon and Nobiin
Bonaventure F. P. Dossou, Mohammed Sabry
Bonaventure F. P. Dossou, Mohammed Sabry
EACL 2021
Crowdsourced Phrase-Based Tokenization for Low-Resourced Neural Machine Translation: The case of Fon Language
Bonaventure F. P. Dossou, Chris C. Emezue
Bonaventure F. P. Dossou, Chris C. Emezue
ACL 2020
Awards & Honours
Selected awards, grants, honours, and research service from my CV.
2025
Global Recognition Award of Science
International recognition received in 2025 for scientific work and impact.
2025
Constructor University Alumni Innovator Award
Alumni innovation recognition from Constructor University.
2024
Borealis AI PhD Fellowship Award
PhD fellowship recognition from RBC Borealis AI supporting machine learning research.
2024
United Nations University Scholars Leadership Symposium
Selected as a delegate for the United Nations University Scholars Leadership Symposium.
2024
Nightingale Tuberculosis Challenge
Honorable mention for a solution focused on detecting active tuberculosis bacilli.
2023
Deep Learning Indaba Best Poster Awards
Received two Best Poster Awards at the 2023 Deep Learning Indaba.
2023
Mila Impact Annual Report
Research featured in Mila's 2023 Impact Annual Report.
2022, 2023
Nightingale High-Risk Breast Cancer Challenge
Winning solutions in Nightingale challenges focused on predicting high-risk breast cancer.
2022
Mila Impact Annual Report
Research featured in Mila's 2022 Impact Annual Report.
2023
McGill Engineering Doctoral Award
Doctoral research award from McGill Engineering.
2022
German African Diaspora Innovation Award
Innovation recognition from the German African diaspora.
2022
Dean's Prize for Outstanding Master's Thesis
Awarded by Jacobs University for an outstanding master's thesis.
2021
Shuttleworth Flash Grant
Flash grant supporting innovative technology and public-interest work.
2021
Wikimedia Foundation Research of the Year
Research recognition from the Wikimedia Foundation.
2021
ViVaTech-UNESCO Challenge Winner
Winner of the challenge on cracking language barriers through data and AI.
2021
Jacobs University Community Award
Recognition for innovation, cultural understanding, and diversity.
2021
Lacuna Fund Grant
Research grant supporting Named Entity Recognition for Fon.
2020 to present
Research Reviewing & Service
Reviewing and organizing service across leading machine learning and NLP conferences.
2019
National Russian AI Hackathon
Winner of the National Russian AI Hackathon.
Technical & Language Skills
Programming & Data
Advanced: Python, Java, C, PHP, C#, R for data analytics, and MySQL for databases. Scientific paper experience.
Languages
Native: French, Fon, Goun. Advanced: English. Upper-intermediate: Russian. Beginner: German.