[{"data":1,"prerenderedAt":383},["ShallowReactive",2],{"work-collection":3},[4,27,45,63,76,92,112,127,139,155,169,181,192,205,217,228,239,250,261,273,287,298,313,323,338,346,354,365,374],{"id":5,"title":6,"url":7,"description":8,"status":9,"tags":10,"authors":16,"venue":22,"featured":23,"kind":24,"type":25,"searchText":26},"paper:cross-model-generalization-of-mechanistic-interpretability-probes","Cross-Model Generalization of Mechanistic-Interpretability Probes","/research/cross-model-generalization-of-mechanistic-interpretability-probes","A reproduction-first audit across model families, with probes refit for each target model.","Manuscript",[11,12,13,14,15],"Mechanistic Interpretability","Activation Probes","Generalization","Reproducibility","Replication",[17,18,19,20,21],"Baimam Boukar","Johannes Taraz","Florent Draye","Terry Jingchen Zhang","Zhijing Jin","",true,"refit","paper","cross-model generalization of mechanistic-interpretability probes a reproduction-first audit of activation-probing methods across model families. we examine how response generation, activation-site selection, probe design, and operating thresholds affect the scope of generalization claims. each target model receives a newly fitted probe; the study does not test zero-shot transfer of probe weights. baimam boukar johannes taraz florent draye terry jingchen zhang zhijing jin manuscript mechanistic interpretability activation probes generalization reproducibility replication cross-model generalization of mechanistic-interpretability probes a reproduction-first audit of activation-probing methods across model families",{"id":28,"title":29,"url":30,"description":31,"status":32,"tags":33,"authors":40,"venue":22,"featured":23,"kind":43,"type":25,"searchText":44},"paper:selection-regret-auditing-deception-probes-as-intervention-selectors","Selection Regret: Auditing Deception Probes as Intervention Selectors","/research/selection-regret-auditing-deception-probes-as-intervention-selectors","Testing whether a probe that detects deception can also select an intervention that changes behavior.","Under Review",[11,34,35,36,37,38,39],"Deception Monitoring","Activation Steering","Evaluation","Linear Probing","Behavioral Evaluation","AI Safety",[17,41,42],"Pauline Nyaboe","Allassan A. Nken","selection","selection regret: auditing deception probes as intervention selectors we test whether accurate deception probes can select useful interventions. the audit compares probe-derived and contrastive steering under held-out behavioral tests and explicit limits on capability loss. detection accuracy and intervention quality require separate evidence. the experiments use controlled, task-defined deceptive choices and do not establish intent or a deployable intervention selector. baimam boukar pauline nyaboe allassan a. nken under review mechanistic interpretability deception monitoring activation steering evaluation linear probing behavioral evaluation ai safety selection regret: auditing deception probes as intervention selectors testing whether accurate deception probes can select useful interventions",{"id":46,"title":47,"url":48,"description":49,"status":50,"tags":51,"authors":55,"venue":22,"featured":60,"kind":61,"type":25,"searchText":62},"paper:phoenix-safe-github-issue-resolution-via-multi-agent-llms","Phoenix: Safe GitHub Issue Resolution via Multi-Agent LLMs","/research/phoenix-safe-github-issue-resolution-via-multi-agent-llms","Multi-agent GitHub issue resolution with baseline-aware safety checks","Preprint",[52,39,53,36,54],"LLM Agents","Safety Gates","Static Analysis",[56,57,58,59],"Kipngeno Koech","Muhammad Adam","Baimam Boukar Jean Jacques","Joao Barros",false,null,"phoenix: safe github issue resolution via multi-agent llms autonomous llm agents can synthesize code changes at scale, but deploying them safely on real codebases requires guarantees that generated patches preserve correctness. phoenix is a multi-agent llm system that resolves github issues from triage through pull-request creation, combining seven layered safety controls with a baseline-aware test evaluation strategy. work is decomposed across six specialized agents — planner, reproducer, coder, tester, failure analyst, and pr agent — coordinated by a label-based github webhook state machine, with every change checked against a baseline test run before a pull request is opened. kipngeno koech muhammad adam baimam boukar jean jacques joao barros preprint llm agents ai safety safety gates evaluation static analysis phoenix multi-agent github issue resolution with baseline-aware safety checks",{"id":64,"title":65,"url":66,"description":22,"status":50,"tags":67,"authors":72,"venue":22,"featured":60,"kind":61,"type":25,"searchText":75},"paper:zero-shot-neural-priors-for-generalizable-cross-subject-and-cross-task-eeg-decoding","Zero-Shot Neural Priors for Generalizable Cross-Subject and Cross-Task EEG Decoding","/research/zero-shot-neural-priors-for-generalizable-cross-subject-and-cross-task-eeg-decoding",[68,69,70,71],"EEG","Brain-Computer Interfaces","Zero-Shot Learning","Neural Decoding",[58,73,74,41,56],"Brandone Fonya","Nchofon Tagha Ghogomu","zero-shot neural priors for generalizable cross-subject and cross-task eeg decoding the development of generalizable electroencephalography (eeg) decoding models is essential for robust brain-computer interfaces (bci) and objective neural biomarkers in mental health. conventional approaches have been hindered by poor cross-subject and cross-task generalization, owing to high inter-subject variability and non-stationary neural signals. we address this challenge with a zero-shot cross-subject decoding framework on the large-scale healthy brain network dataset,benchmarking a convolutional neural network baseline, a hybrid lstm, and a transformer-based foundation model. to adapt the transformer for regression while averting catastrophic forgetting, we propose a novel progressive unfreezing strategy. the baseline yielded an nrmse of 0.9991, whereas our fine-tuned transformer achieved 0.9799 on unseen subjects.this work establishes scalable, calibration-free eeg decoding for computational psychiatry and behavioral prediction. baimam boukar jean jacques brandone fonya nchofon tagha ghogomu pauline nyaboe kipngeno koech preprint eeg brain-computer interfaces zero-shot learning neural decoding",{"id":77,"title":78,"url":79,"description":22,"status":50,"tags":80,"authors":85,"venue":90,"featured":60,"kind":61,"type":25,"searchText":91},"paper:retrieval-with-multiple-query-vectors-through-anomalous-pattern-detection","Retrieval with Multiple Query Vectors through Anomalous Pattern Detection","/research/retrieval-with-multiple-query-vectors-through-anomalous-pattern-detection",[81,82,83,84],"Information Retrieval","DeepScan","LLM Embeddings","Anomalous Pattern",[86,58,87,88,89],"Allassan Tchangmena A Nken","Miriam Rateike","Celia Cintas","Skyler Speakman","AAAI 2026 Workshop on New Frontiers in Information Retrieval","retrieval with multiple query vectors through anomalous pattern detection a classical vector retrieval problem typically considers a single query embedding vector as input and retrieves the most similar embedding vectors from a vector database. however, complex reasoning and retrieval tasks frequently require multiple query vectors, rather than a single one. in this work, we propose a retrieval method that considers multiple query vectors simultaneously and retrieves the most relevant vectors from the database using concepts from anomalous pattern detection. specifically, our approach leverages a set of query vectors q (with |q| ≥ 1), and identifies the subset of vector dimensions within q that standout (anomalous) from the rest of dimensions. next, we scan the vector database to retrieve the set of vectors that are also anomalous across the previously identified vector dimensions and return them as our retrieved set of vectors. we validate our approach on two image datasets, a text dataset, and a tabular dataset. allassan tchangmena a nken baimam boukar jean jacques miriam rateike celia cintas skyler speakman preprint aaai 2026 workshop on new frontiers in information retrieval information retrieval deepscan llm embeddings anomalous pattern",{"id":93,"title":94,"url":95,"description":96,"status":97,"tags":98,"authors":105,"venue":110,"featured":60,"kind":61,"type":25,"searchText":111},"paper:mapping-socioeconomic-air-quality-disparities-in-rwanda-using-sentinel-5p-tropomi-data-in-google-earth-engine","Mapping Socioeconomic Air Quality Disparities In Rwanda Using Sentinel-5P TROPOMI Data In Google Earth Engine","/research/mapping-socioeconomic-air-quality-disparities-in-rwanda-using-sentinel-5p-tropomi-data-in-google-earth-engine","Notebooks relating Sentinel-5P air-quality measurements to socioeconomic data in Rwanda using Google Earth Engine.","Published",[99,100,101,102,103,104],"Air Quality","Remote Sensing","Google EE","Python","Google Earth Engine","Geospatial analysis",[58,106,107,108,109],"Kamikazi Ndayizeye Raissa","Umubyeyi Evelyne","Bertinn Ndahayo Singizwa","Emily Aiken","IEEE MIGARS 2025","mapping socioeconomic air quality disparities in rwanda using sentinel-5p tropomi data in google earth engine this study investigates socioeconomic disparities in air quality across rwanda using sentinel-5p tropomi satellite data processed through google earth engine. we develop novel methodologies to correlate atmospheric pollutant concentrations with demographic and economic indicators, revealing significant patterns in environmental justice. our analysis demonstrates how satellite-based monitoring can inform policy decisions for equitable environmental health outcomes. baimam boukar jean jacques kamikazi ndayizeye raissa umubyeyi evelyne bertinn ndahayo singizwa emily aiken published ieee migars 2025 air quality remote sensing google ee python google earth engine geospatial analysis rwanda air quality analysis notebooks relating sentinel-5p air-quality measurements to socioeconomic data in rwanda using google earth engine.",{"id":113,"title":114,"url":115,"description":116,"status":117,"tags":118,"authors":124,"venue":125,"featured":60,"kind":61,"type":25,"searchText":126},"paper:explainable-deep-learning-based-potentially-hazardous-asteroids-classification-using-graph-neural-networks","Explainable Deep-Learning Based Potentially Hazardous Asteroids Classification Using Graph Neural Networks","/research/explainable-deep-learning-based-potentially-hazardous-asteroids-classification-using-graph-neural-networks","Graph neural network experiments for classifying potentially hazardous asteroids using NASA JPL small-body data.","Presented",[119,120,121,102,122,123],"Space","AI","Astrophysics","Graph neural networks","Astronomy",[58],"Harvard AstroAI Workshop 2025.","explainable deep-learning based potentially hazardous asteroids classification using graph neural networks classifying potentially hazardous asteroids (phas) is crucial for planetary defense. while the official designation relies on simple metrics (earth moid and absolute magnitude h), we explore if complex dynamical relationships offer a robust, interpretable hazard assessment framework. we introduce a graph neural network (gnn) approach that models asteroids as nodes defined by 11 orbital and physical features. edges are defined using the tisserand parameter relative to jupiter (tj) as a proxy for dynamical similarity, allowing the gnn to capture latent relationships among asteroid groups. on a dataset with a 0.22% hazardous class, the paper reports 99% accuracy and 0.99 auroc, alongside 78% recall and 37% f1 for hazardous asteroids after smote. baimam boukar jean jacques presented harvard astroai workshop 2025. space ai astrophysics python graph neural networks astronomy asteroid classification with gnns graph neural network experiments for classifying potentially hazardous asteroids using nasa jpl small-body data.",{"id":128,"title":129,"url":130,"description":22,"status":97,"tags":131,"authors":134,"venue":110,"featured":60,"kind":61,"type":25,"searchText":138},"paper:humidity-inference-with-geographic-features-and-machine-learning-for-enhanced-contrail-prediction-for-african-airspace","Humidity Inference with Geographic Features and Machine Learning for Enhanced Contrail Prediction for African Airspace","/research/humidity-inference-with-geographic-features-and-machine-learning-for-enhanced-contrail-prediction-for-african-airspace",[120,132,133],"Aviation","Contrails",[135,58,136,137],"Alice Mugengano","Jonathan Kayizzi","Moise Busogi","humidity inference with geographic features and machine learning for enhanced contrail prediction for african airspace air travel's 3.8% yearly growth amplifies aviation's impact, with contrails causing about 4% of global warming. in africa, sparse meteorological data and coarse era5 resolution underpredict contrails in cities like nairobi and lagos. this study uses a random forest model within pycontrails to infer upper-tropospheric humidity (rhi) from water proximity, forest density, and topography. using global datasets, it achieves an r2 of 0.59 identifying water proximity as the main driver. african cities show only 32 contrail segments versus 450 in europe, underscoring the model's potential to refine predictions and reduce aviation's non-co₂ climate effects. alice mugengano baimam boukar jean jacques jonathan kayizzi moise busogi published ieee migars 2025 ai aviation contrails",{"id":140,"title":141,"url":142,"description":143,"status":144,"tags":145,"authors":151,"venue":153,"featured":60,"kind":61,"type":25,"searchText":154},"paper:causal-structure-analysis-for-telemetry-anomaly-detection-in-spacecraft-systems","Causal Structure Analysis for Telemetry Anomaly Detection in Spacecraft Systems","/research/causal-structure-analysis-for-telemetry-anomaly-detection-in-spacecraft-systems","Probabilistic graphical models and deep-learning experiments for anomaly detection in ESA-Mission1 spacecraft telemetry.","Ongoing",[146,147,148,102,149,150],"Satellites Operations","Telemetry","Graphical Models","Anomaly detection","Probabilistic models",[58,152],"Kip Koech","10th TTC International Workshop","causal structure analysis for telemetry anomaly detection in spacecraft systems spacecraft telemetry anomaly detection is critical for mission success and safety. this research applies causal inference methods to understand the underlying causal structure in spacecraft telemetry data. we developed graphical models that can identify root causes of anomalies, enabling more effective predictive maintenance and mission planning strategies. baimam boukar jean jacques kip koech ongoing 10th ttc international workshop satellites operations telemetry graphical models python anomaly detection probabilistic models spacecraft telemetry anomaly detection probabilistic graphical models and deep-learning experiments for anomaly detection in esa-mission1 spacecraft telemetry.",{"id":156,"title":157,"url":158,"description":159,"status":22,"tags":160,"authors":166,"venue":22,"featured":60,"kind":61,"type":167,"searchText":168},"project:/projects/anora","Anora AI","/projects/anora","Make sense of your enterprise data like never before",[161,162,163,164,165],"Flutter","Data","KnowledgeBase","Gemini","Firebase",[],"project","anora ai make sense of your enterprise data like never before flutter data knowledgebase gemini firebase",{"id":170,"title":171,"url":172,"description":173,"status":174,"tags":175,"authors":178,"venue":22,"featured":23,"kind":179,"type":167,"searchText":180},"project:/projects/coordinate-or-concept-mismatch","Coordinate or Concept Mismatch?","/projects/coordinate-or-concept-mismatch","Investigating cross-model probe-transfer failures and task-conditioned alignment","Ongoing Research",[11,176,177,13],"Probe Transfer","Representation Alignment",[],"alignment","coordinate or concept mismatch? investigating cross-model probe-transfer failures and task-conditioned alignment ongoing research mechanistic interpretability probe transfer representation alignment generalization",{"id":182,"title":183,"url":184,"description":185,"status":186,"tags":187,"authors":190,"venue":22,"featured":60,"kind":61,"type":167,"searchText":191},"project:/projects/cosmosense","Cosmosense","/projects/cosmosense","A mobile app to keep space enthusiasts passionate","Live",[161,119,188,189],"NASA","SpaceX",[],"cosmosense a mobile app to keep space enthusiasts passionate live flutter space nasa spacex",{"id":193,"title":194,"url":195,"description":196,"status":22,"tags":197,"authors":203,"venue":22,"featured":60,"kind":61,"type":167,"searchText":204},"project:/projects/deepface","Deep Face Verification with CNNs","/projects/deepface","Deep Face Classification, and Verification",[198,199,200,201,202],"PyTorch","Metric Learning","TripletLoss","ArcFace","Retrieval",[],"deep face verification with cnns deep face classification, and verification pytorch metric learning tripletloss arcface retrieval",{"id":206,"title":207,"url":208,"description":209,"status":210,"tags":211,"authors":215,"venue":22,"featured":60,"kind":61,"type":167,"searchText":216},"project:/projects/deepscanner","IBM DeepScanner","/projects/deepscanner","Research infrastructure for analyzing anomalous patterns in LLM activations","Internal Tool",[102,198,212,213,214],"Ray","IBM Cloud","LLMs",[],"ibm deepscanner research infrastructure for analyzing anomalous patterns in llm activations internal tool python pytorch ray ibm cloud llms",{"id":218,"title":219,"url":220,"description":221,"status":174,"tags":222,"authors":225,"venue":22,"featured":23,"kind":226,"type":167,"searchText":227},"project:/projects/delusion-reinforcement","Measuring Delusion Reinforcement Across the Directness-Need Spectrum","/projects/delusion-reinforcement","Separating affective support from epistemic sycophancy in conversational models",[223,224,38],"LLM Safety","Epistemic Sycophancy",[],"support","measuring delusion reinforcement across the directness-need spectrum separating affective support from epistemic sycophancy in conversational models ongoing research llm safety epistemic sycophancy behavioral evaluation",{"id":229,"title":230,"url":231,"description":232,"status":22,"tags":233,"authors":237,"venue":22,"featured":60,"kind":61,"type":167,"searchText":238},"project:/projects/ecocaasitech","EcoCaasitech","/projects/ecocaasitech","Collect and recycle waste for a cleaner environment.",[161,234,235,236],"Waste Collection","Recycling","EcoPoints",[],"ecocaasitech collect and recycle waste for a cleaner environment. flutter waste collection recycling ecopoints",{"id":240,"title":241,"url":242,"description":243,"status":22,"tags":244,"authors":248,"venue":22,"featured":60,"kind":61,"type":167,"searchText":249},"project:/projects/electchain","Electchain","/projects/electchain","A SaaS mobile voting app for transparent and secure elections",[161,245,246,247,165],"Blockchain","Voting","SaaS",[],"electchain a saas mobile voting app for transparent and secure elections flutter blockchain voting saas firebase",{"id":251,"title":252,"url":253,"description":254,"status":22,"tags":255,"authors":259,"venue":22,"featured":60,"kind":61,"type":167,"searchText":260},"project:/projects/elite","Elite","/projects/elite","Stay sync with Cameroonian Football Events",[161,256,257,258,165],"Football","EliteOne","Cameroon",[],"elite stay sync with cameroonian football events flutter football eliteone cameroon firebase",{"id":262,"title":263,"url":264,"description":265,"status":186,"tags":266,"authors":271,"venue":22,"featured":60,"kind":61,"type":167,"searchText":272},"project:/projects/intellibra","Intellibra","/projects/intellibra","AI-powered thermography system for breast cancer screening",[267,268,269,270,102],"Vision-Language Models","Edge AI","Deep Learning","Medical Imaging",[],"intellibra ai-powered thermography system for breast cancer screening live vision-language models edge ai deep learning medical imaging python",{"id":274,"title":275,"url":276,"description":277,"status":278,"tags":279,"authors":285,"venue":22,"featured":60,"kind":61,"type":167,"searchText":286},"project:/projects/limasense-mlops","LimaSense MLOps","/projects/limasense-mlops","Production ML backend for crop detection and quality grading","Production System",[280,281,282,283,284],"FastAPI","YOLO","Supabase","Docker","AWS App Runner",[],"limasense mlops production ml backend for crop detection and quality grading production system fastapi yolo supabase docker aws app runner",{"id":288,"title":289,"url":290,"description":291,"status":292,"tags":293,"authors":295,"venue":22,"featured":23,"kind":296,"type":167,"searchText":297},"project:/projects/matryoshka-saes","Replicating Matryoshka Sparse Autoencoders","/projects/matryoshka-saes","Replicating feature recovery and comparing Matryoshka and standard sparse autoencoders","Published Writeup (LessWrong)",[11,294,198,15],"Sparse Autoencoders",[],"sae","replicating matryoshka sparse autoencoders replicating feature recovery and comparing matryoshka and standard sparse autoencoders published writeup (lesswrong) mechanistic interpretability sparse autoencoders pytorch replication",{"id":299,"title":300,"url":301,"description":302,"status":303,"tags":304,"authors":311,"venue":22,"featured":60,"kind":61,"type":167,"searchText":312},"project:/projects/mytorch","MyTorch","/projects/mytorch","A foundational deep learning library built from first principles in Python","Active Development",[305,269,198,102,306,307,308,309,310],"Neural Networks","C++","MLPs","CNNs","RNNs","Transformers",[],"mytorch a foundational deep learning library built from first principles in python active development neural networks deep learning pytorch python c++ mlps cnns rnns transformers",{"id":314,"title":315,"url":316,"description":317,"status":22,"tags":318,"authors":321,"venue":22,"featured":60,"kind":61,"type":167,"searchText":322},"project:/projects/sos","SOSbyCaasitech","/projects/sos","Stay Safe, Stay Connected.",[161,319,320],"Safety","Emergency",[],"sosbycaasitech stay safe, stay connected. flutter safety emergency",{"id":324,"title":325,"url":326,"description":327,"status":328,"tags":329,"authors":336,"venue":22,"featured":60,"kind":61,"type":167,"searchText":337},"project:/projects/sso-platform","Rwanda National SSO","/projects/sso-platform","A sovereign micro services based Single Sign-On system for RISA.","Proof-of-Concept",[330,280,331,332,333,334,102,335],"Keycloak","Next.js","SSO","IAM","Authentication","Node.js",[],"rwanda national sso a sovereign micro services based single sign-on system for risa. proof-of-concept keycloak fastapi next.js sso iam authentication python node.js",{"id":339,"title":340,"url":341,"description":342,"status":22,"tags":343,"authors":344,"venue":22,"featured":60,"kind":61,"type":167,"searchText":345},"project:https://github.com/baimamboukar/DotNetPerfMonitor",".NETPerfMonitor","https://github.com/baimamboukar/DotNetPerfMonitor","A performance regression benchmarking tool for .NET ecosystem",[],[],".netperfmonitor a performance regression benchmarking tool for .net ecosystem",{"id":347,"title":348,"url":349,"description":350,"status":22,"tags":351,"authors":352,"venue":22,"featured":60,"kind":61,"type":167,"searchText":353},"project:https://github.com/baimamboukar/card_game","BlackMoon","https://github.com/baimamboukar/card_game","Black Jack game in C++ OOP",[],[],"blackmoon black jack game in c++ oop",{"id":355,"title":356,"url":357,"description":358,"status":22,"tags":359,"authors":363,"venue":22,"featured":60,"kind":61,"type":167,"searchText":364},"project:https://github.com/baimamboukar/figure-studio","Figure Studio","https://github.com/baimamboukar/figure-studio","A browser editor for charts, annotated figures, and multi-panel layouts, with PNG, JPEG, and SVG export.",[360,361,362],"Vue","Data visualization","Research tools",[],"figure studio a browser editor for charts, annotated figures, and multi-panel layouts, with png, jpeg, and svg export. vue data visualization research tools",{"id":366,"title":367,"url":368,"description":369,"status":22,"tags":370,"authors":372,"venue":22,"featured":60,"kind":61,"type":167,"searchText":373},"project:https://github.com/baimamboukar/paperflow","PaperFlow","https://github.com/baimamboukar/paperflow","Python tools for converting LaTeX papers into web pages with equations, figures, citations, and cross-references.",[102,371,362],"LaTeX",[],"paperflow python tools for converting latex papers into web pages with equations, figures, citations, and cross-references. python latex research tools",{"id":375,"title":376,"url":377,"description":378,"status":22,"tags":379,"authors":381,"venue":22,"featured":60,"kind":61,"type":167,"searchText":382},"project:https://github.com/baimamboukar/autoregressive-language-modeling-with-transformer-decoder","Transformer Sequence Modeling","https://github.com/baimamboukar/autoregressive-language-modeling-with-transformer-decoder","PyTorch implementations of decoder-only language modeling and encoder-decoder speech recognition, with training and decoding pipelines.",[198,310,380],"Speech recognition",[],"transformer sequence modeling pytorch implementations of decoder-only language modeling and encoder-decoder speech recognition, with training and decoding pipelines. pytorch transformers speech recognition",1791285324456]