https://taimaka.org/blog daily 0.75 2026-10-29 https://taimaka.org/blog/technical-overview-leveraging-machine-learning-to-enhance-patient-outcomes-in-malnutrition-care monthly 0.5 2026-10-29 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/fc9767bc-b0af-48a0-9278-44577c23af2e/Image+10-29-26+at+2.30%E2%80%AFPM.jpg Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 1 - Events https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/2ca93c83-4c18-4def-8814-da3f59f24650/Screenshot+2026-10-29+at+2.41.37%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 2 - Patient enrollment by length of stay https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/960f4e33-9742-4146-acd4-a18f36b69c77/Screenshot+2026-10-29+at+4.42.19%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 3 - Final MUAC by length of stay and discharge status https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/7295cd43-7f21-40e3-8cf3-43b25cf957f8/Screenshot+2026-10-29+at+4.49.46%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 4 - Discharge reason by length of stay in program https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/6880df9a-994c-48b7-9d2e-7482c6cb86d0/Screenshot+2026-10-29+at+4.50.56%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 5 - Scatter plot of length of stay and MUAC at week 1, color-coded by mortality status https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/8bce381a-7365-4275-b09b-5617cc738017/Screenshot+2026-10-29+at+4.53.44%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 6 - Precision recall by cutoff, stacked survival https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/d0626c37-30d3-49f5-b7d0-3dc8ffe5cf0a/Screenshot+2026-10-29+at+4.55.26%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Precision recall by cutoff, classification https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/d9db2773-5406-47b8-9820-014c3d91709b/Screenshot+2026-10-29+at+4.57.18%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 8 - AutoML steps https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c425ef84-b1d4-4beb-ba08-cc55e7b8f277/Screenshot+2026-10-29+at+4.59.26%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 9 - Feature reduction and model performance https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/d71f4698-3cd4-4714-842a-3d06de54a5d0/Screenshot+2026-10-29+at+5.26.28%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 10 - Number of models per event based on stratification cohorts https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/354b996c-be42-4482-b212-04650165a4ae/Screenshot+2026-10-29+at+5.15.49%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 11 - Precision-recall curve cutoff choice https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e012b63c-1a2b-4760-a927-e09ccc4307c9/Screenshot+2026-10-29+at+5.18.50%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 12 - SHAP summary (“beeswarm”) plot of model feature importance https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/22e4559f-31b8-4269-b079-fa318ae1567a/Screenshot+2026-10-29+at+5.27.38%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 13 - SHAP waterfall plot of feature importance for a specific patient https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/d0e00d28-4e5e-4674-9b7c-2ff14aa88aaa/Screenshot+2026-10-29+at+5.21.41%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 14 - Weight prediction for the next 30 days with 90% confidence range in blue and hover text displaying for July 11, 2026 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/bee5ec81-8942-43b4-b966-0e57596de375/Screenshot+2026-10-29+at+5.23.02%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 15 - Weibull parameter values, with rho highlighted https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e87396ac-7599-4b27-99fc-dba3b351079c/Screenshot+2026-10-29+at+5.24.07%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 16 - Survival probability over time for a patient https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/f546a3c3-8a66-4db2-aaf6-940bef2ebccb/Screenshot+2026-10-29+at+5.25.12%E2%80%AFPM.png Blog - Technical Overview: Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care Figure 17 - Systems and data flow https://taimaka.org/blog/leveraging-machine-learning-to-enhance-patient-outcomes-in-malnutrition-care monthly 0.5 2026-10-29 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/10e1b300-9bcb-4ae7-89d9-d28da93c679d/exported_6CADF65B-B9C9-4673-9614-81B6D7F81F0A.JPG Blog - Leveraging Machine Learning to Enhance Patient Outcomes in Malnutrition Care https://taimaka.org/blog/patient-identification-system monthly 0.5 2026-03-13 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/36ff48c0-b684-4ce6-aa3a-64fbe0c1bf3f/facial_recognition_system.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out An example of a facial recognition system. The landmarking model identifies locations of important facial features like eyes, nose, mouth, etc. Using these landmarks, the image is transformed and cropped and then passed to the embedding model, which creates the embedding vector for each image. Finally, the two embeddings are compared. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/3cb0b322-1caf-4eab-a65a-87149428fbe1/embedding_similarity_histogram.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Similarity scores based on embeddings created by FaceNet. The orange histogram represents the distribution of similarity scores computed on two photos of different patients, and the blue histogram represents the distribution of similarity scores of two different photos of the same patient. There is significant overlap of these two distributions which implies that false rejection and acceptance rates would be too high to be useful. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/0ce11169-0540-4187-a173-efb668c4556c/model_architecture.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out An illustration of the architecture of our FaceNet model. The image of a patient’s face first passes through the base FaceNet architecture which outputs a 512-dimensional embedding. That embedding then passes through a custom head consisting of two layers. The final output is a 128-dimensional embedding. (This diagram is for illustration only; the actual number of FaceNet layers is much larger.) https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e6267824-4a87-4dd4-bd94-46cebd0981b3/embedding_training_visualization.gif Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out A visualization of face embeddings in three dimensions. Vectors that end in a dot are embeddings of a single photo, and vectors that end in a diamond are the average embedding of all photos for a given patient. As model training progresses, embeddings of different patients are pushed further apart. Note that the actual embeddings used in our system lie on a 128-dimensional hypersphere, rather than the 3-dimensional sphere used for this visualization. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/07ca8911-d5ec-4d11-9012-aa5b7bf9cac1/cosine_margin_triplet_loss.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out The formula for the cosine-margin-triplet loss. θ+ is the angle between the anchor and comparison embedding, and θ- is the angle between the anchor and negative embedding. The hyperparameters s and m are the radius and margin, respectively. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/5d0f0392-a148-4d5a-a3ac-492e49d74e64/cosine_margin_triplet_loss_rewritten.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out The original formula is rewritten here in order to help the reader understand how it works. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/cbb580b7-0acb-40dd-bc8d-1dde59cde341/validation_loss_traceplot.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Cosine-margin-triplet loss on the validation dataset over each phase of training. The epoch with the best loss is marked with a vertical dashed line for each phase. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/58e24c6b-1059-4a71-aa57-5b1cc4b6637d/adapted_embedding_similarity_histogram.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Similarity scores based on embeddings created by our new version of FaceNet that has been trained on Taimaka patient photos. There is now considerably more separation between the two distributions. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/a4da1279-dc56-43b0-96ab-5bb330eb1186/aggregated_embedding_similarity_histogram.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Similarity scores based on averaged embeddings. The separation has further improved. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c1fc25a6-9392-458b-afae-5d529733aff3/embedding_aggregation_performance_analysis.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Performance as a function of the recency weight w. The solid lines are the metrics for EWMA templates, and the dashed lines are the metrics for a simple average template, provided for comparison. The blue, orange, and green lines are true rejection rate (TRR), true acceptance rate (TAR), and top 3 accuracy, respectively. While a recency weight of around 0.4 is optimal, it is not better than a simple average template. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/64c55e3b-297b-4165-96cf-4c41985eaa65/embedding_similarity_performance_over_time.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out Facial similarity scores for 709 photos of 102 patients. The cluster of points on the left are the photos from the original visits, and the cluster on the right are the photos from the checkback survey. A trend line and 95% prediction band are shown, and the horizontal dashed line is the cutoff used for matching. https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e6594acf-e66f-421e-997d-d27f55509271/acceptance_rate_performance_over_time.png Blog - Taimaka ID: Building a patient identification system for pediatric malnutrition treatment in Nigeria - Make it stand out True acceptance rate of returning patients as a function of time. The true acceptance rate is calculated from the fitted trend in the previous figure using the proportion of the estimated distribution that lies above the matching cutoff of 0.8. Performance degrades significantly when the lag between visits exceeds five months. https://taimaka.org/taimaka-home daily 1.0 2026-07-25 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/42dbf7e6-d323-4163-85fc-780896fc9ea8/1280px-Founders_Pledge_logo.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/254d8d8a-4abd-4460-89ca-e08ebac29fd5/GW_Logo_Standard_300ppi_CMYK.jpg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/030dd541-ca42-426a-b5c2-ea1a42390638/Screenshot_2025-04-24_at_3.04.20_PM-removebg-preview.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e3bfa62c-7001-47b5-a0f1-af70ed77a4ae/Dovetail_logo_horizontal_RGB_2col_large.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/fe593b3e-c154-4273-93be-cf3c9d9a8f2b/Screenshot+2025-04-24+at+3.42.41%E2%80%AFPM.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e20569e8-3896-4748-9743-e078c8d8e94d/Screenshot+2025-04-24+at+3.44.11%E2%80%AFPM.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/eed7a3b3-07e2-4aff-b42f-f797fa683df1/DSC_0476.JPG https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/0cfbe455-e158-4657-896a-4a008fe7420c/DSC_0506.jpg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/2853d21a-61b0-4122-9835-5eafd945cc0e/DSC_0379.jpeg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/0428943e-9db8-48fc-8900-1dce060b6f2b/DSC_0391.JPG https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/3a26d41f-aa8f-41e9-9994-c7813a9f2a38/DSC_0075-EDIT.jpg https://video.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/22290a82-aeb6-47eb-ae4d-1520df948a42/playlist.m3u8 https://video.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/22290a82-aeb6-47eb-ae4d-1520df948a42/thumbnail Home Home https://taimaka.org/model daily 0.75 2025-06-10 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/8a8685d3-07d4-4056-bd68-396b5ac7e72c/image.jpg Model https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/a78e440d-c631-498f-a352-0ea7f9c076e9/DSC_0433.JPG Model https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/f2100645-6544-4262-81dd-7a56dd186be0/7DBA0943-1EEE-4446-9781-967FD68DE47C_1_105_c.jpeg Model https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/e6ac5394-22dc-4dfe-9636-aa5434088125/Screenshot+2024-08-23+at+3.51.02%E2%80%AFPM.png Model https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/b65834b5-24f5-4d3f-91a2-4fb4d69f4afd/DSC_0136.JPG https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/eed7a3b3-07e2-4aff-b42f-f797fa683df1/DSC_0476.JPG https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/1689622698739-BQ96FWXR9QES7SEABS8S/DSC_0557.JPG https://taimaka.org/impact daily 0.75 2025-10-17 https://images.squarespace-cdn.com/content/v1/5ec321c2af33de48734cc929/1618511387030-5LI1E5QMVTQ2RY9S00A4/20140228_Trade+151_0046+1.jpg Impact https://images.squarespace-cdn.com/content/v1/5ec321c2af33de48734cc929/1618511440608-5BVMHVN4WR2I9YQV1WWR/Aro+Ha_0010+1.jpg Impact https://images.squarespace-cdn.com/content/v1/5ec321c2af33de48734cc929/1618511467879-K5J9BYNACHJVDV3BNSFU/Aro+Ha_0387+1.jpg Impact https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/3b733d18-e6fa-4e27-9d75-d2d4ac46833a/Screenshot+2024-08-14+at+4.35.56%E2%80%AFPM.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/1496c5d6-be51-4220-8177-5c6a5dd4a81d/Screenshot+2024-08-14+at+4.35.22%E2%80%AFPM.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/00d71416-a464-4e25-b052-f650140855d0/Screenshot+2024-08-14+at+4.37.40%E2%80%AFPM.png https://taimaka.org/who-we-are daily 0.75 2026-03-31 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/dca39f7e-53a4-42f9-afb9-aade25bc5734/Screen+Shot+2023-05-24+at+11.59.47+AM.png Who We Are https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/1663934107857-ICEU1QS9LD969YSLSKH4/DSC_0306.jpeg Who We Are https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c7b1f6e1-0243-467f-8859-41eb3a425846/DSC_0008.JPG Who We Are https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c0b9050c-76de-40a7-8fb1-b3c5416ec7d0/1_JHppodF8HUJsZaRYG6JT9w.jpeg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/1628893429958-88ZDO75DCODPR57DORHG/c7cc026c-28a1-4970-bbf2-430e4c104aa6.jpg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c2e6fe67-4987-4e43-9255-19f05e709604/DSC_0272+2.JPG https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/3cc2e836-c393-4db3-931d-465bdcffcb99/Screenshot+2024-08-15+at+2.35.07%E2%80%AFPM.png https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/10558471-42b5-49df-b481-0da5fd328cac/CE+Headshot.jpg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/14d3e26b-0395-47a9-874c-cfd900b401e4/img_9253_720.jpg https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c00453e8-fc28-4cf3-ac75-123088bcd521/DSC_0308+2.JPG https://taimaka.org/donate daily 0.75 2026-07-25 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/bfc4142d-094b-47f2-befc-bd9121f9ae61/DSC_0052.jpeg https://taimaka.org/work-with-us daily 0.75 2026-04-09 https://images.squarespace-cdn.com/content/v1/5d3e07692e027a00013be3f4/c7b1f6e1-0243-467f-8859-41eb3a425846/DSC_0008.JPG