<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Grace UMUTONIWASE</title>
    <description>The latest articles on DEV Community by Grace UMUTONIWASE (@grace_umutoniwase_b6e978d).</description>
    <link>https://dev.to/grace_umutoniwase_b6e978d</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4154171%2Ffa2fc0ea-9dd8-4442-9642-221ddb15fac4.jpeg</url>
      <title>DEV Community: Grace UMUTONIWASE</title>
      <link>https://dev.to/grace_umutoniwase_b6e978d</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/grace_umutoniwase_b6e978d"/>
    <language>en</language>
    <item>
      <title>Open-Source Large Language Models: Finding the Best Model for Our Project</title>
      <dc:creator>Grace UMUTONIWASE</dc:creator>
      <pubDate>Mon, 26 Oct 2026 14:45:07 +0000</pubDate>
      <link>https://dev.to/grace_umutoniwase_b6e978d/open-source-large-language-models-finding-the-best-model-for-our-project-5dfg</link>
      <guid>https://dev.to/grace_umutoniwase_b6e978d/open-source-large-language-models-finding-the-best-model-for-our-project-5dfg</guid>
      <description>&lt;h1&gt;
  
  
  Which Open-Source LLM Should We Use to Predict the Next Word?
&lt;/h1&gt;

&lt;p&gt;Our project is about next-word prediction: the model reads the start of a sentence and guesses which token comes next. In our first experiments we opened up GPT-2 (124 million parameters) and also trained a tiny GPT of our own on Kinyarwanda news. One result stood out. GPT-2 found a Kinyarwanda sentence much harder than the same sentence in English, and it needed about 1.7 times as many tokens to write it, because it was trained almost only on English.&lt;/p&gt;

&lt;p&gt;That raised a question for us: is there a newer open-source model that does better, costs nothing, and still runs on a free Google Colab T4 GPU (16 GB)? We looked at five and picked one.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. The models we looked at
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Qwen2.5-1.5B (Alibaba Cloud, Qwen team).&lt;/strong&gt; It has 1.54 billion parameters and a 32,768-token context window. The licence is Apache 2.0 and the model is not gated on Hugging Face, so nobody needs an account or a licence click. It was trained on text in more than 29 languages and is known for following instructions well. There is a base version, which simply continues text (exactly our task), and an Instruct version for chat.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Llama 3.2 3B (Meta).&lt;/strong&gt; It has 3.21 billion parameters and a 128K-token context window. It uses Meta's own Llama 3.2 Community License, and the official files on Hugging Face are gated, so you must log in and accept the licence first. It is a strong all-round small model for chat and summaries. Meta officially supports eight languages, and Kinyarwanda is not one of them.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Phi-3.5-mini (Microsoft).&lt;/strong&gt; It has 3.8 billion parameters and a 128K-token context window, under the very permissive MIT licence. It was trained on carefully chosen, reasoning-heavy data, so it is good at reasoning and long documents for its size. It is the biggest model on our list and focuses mainly on English.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Gemma 3 1B (Google DeepMind).&lt;/strong&gt; It has 1 billion parameters and a 32K-token context window, and it was trained on 2 trillion tokens in over 140 languages with a large 262K-entry vocabulary. It is released under the Gemma Terms of Use, which is not a standard open-source licence, and you must accept those terms on Hugging Face before downloading. It is very light, but at this size it has the least capacity.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;SmolLM2-1.7B (Hugging Face).&lt;/strong&gt; It has 1.7 billion parameters, was trained on 11 trillion tokens, and uses Apache 2.0 with no gating. Its context window is only about 8K tokens, the shortest here, and it is mostly English, so we expect it to struggle with Kinyarwanda just as GPT-2 did.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Comparison table
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Model&lt;/th&gt;
&lt;th&gt;Made by&lt;/th&gt;
&lt;th&gt;Parameters&lt;/th&gt;
&lt;th&gt;Context window&lt;/th&gt;
&lt;th&gt;Licence&lt;/th&gt;
&lt;th&gt;Account or licence click needed?&lt;/th&gt;
&lt;th&gt;Good at&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Qwen2.5-1.5B&lt;/td&gt;
&lt;td&gt;Alibaba (Qwen)&lt;/td&gt;
&lt;td&gt;1.54B&lt;/td&gt;
&lt;td&gt;32K&lt;/td&gt;
&lt;td&gt;Apache 2.0&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;Instructions, many languages, small size&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Llama 3.2 3B&lt;/td&gt;
&lt;td&gt;Meta&lt;/td&gt;
&lt;td&gt;3.21B&lt;/td&gt;
&lt;td&gt;128K&lt;/td&gt;
&lt;td&gt;Llama 3.2 Community&lt;/td&gt;
&lt;td&gt;Yes (gated)&lt;/td&gt;
&lt;td&gt;Chat, summaries, tool calling&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Phi-3.5-mini&lt;/td&gt;
&lt;td&gt;Microsoft&lt;/td&gt;
&lt;td&gt;3.8B&lt;/td&gt;
&lt;td&gt;128K&lt;/td&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;Reasoning, long documents&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gemma 3 1B&lt;/td&gt;
&lt;td&gt;Google DeepMind&lt;/td&gt;
&lt;td&gt;1.0B&lt;/td&gt;
&lt;td&gt;32K&lt;/td&gt;
&lt;td&gt;Gemma Terms of Use&lt;/td&gt;
&lt;td&gt;Yes (gated)&lt;/td&gt;
&lt;td&gt;Light multilingual text&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SmolLM2-1.7B&lt;/td&gt;
&lt;td&gt;Hugging Face&lt;/td&gt;
&lt;td&gt;1.7B&lt;/td&gt;
&lt;td&gt;~8K&lt;/td&gt;
&lt;td&gt;Apache 2.0&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;Short English text&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  3. Our choice: Qwen2.5-1.5B
&lt;/h2&gt;

&lt;p&gt;We chose &lt;strong&gt;Qwen2.5-1.5B&lt;/strong&gt;. Here is how it answers the questions we set ourselves.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does it run on a free Colab T4?&lt;/strong&gt; Yes, with plenty of room. A model needs roughly 2 bytes per parameter in 16-bit precision, so 1.54 billion parameters is about 3 GB. By the same rule, Phi-3.5-mini needs about 7.6 GB and Llama 3.2 3B about 6.5 GB. Both would fit, but with less space left and slower predictions. These are our own estimates, and we will check the real numbers in Colab with &lt;code&gt;torch.cuda.max_memory_allocated()&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Is it free, with no API key or account?&lt;/strong&gt; Yes. It is Apache 2.0 and not gated, so any team member can load it with one line of code. Llama 3.2 and Gemma both need a Hugging Face login and an accepted licence, which is annoying for a group project and for anyone who wants to re-run our notebook.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does it do what next-word prediction needs?&lt;/strong&gt; Yes. Like GPT-2, it is a decoder-only transformer. We feed in tokens, read the probabilities for the next token and compute the loss, so we can reuse the &lt;code&gt;compute_loss&lt;/code&gt; and &lt;code&gt;continue_text&lt;/code&gt; functions we already wrote. Its context window of 32K tokens is also far bigger than GPT-2's 1,024 positions or our tiny GPT's 64.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How fast is it and how much memory does it use?&lt;/strong&gt; At about 3 GB it is the second-lightest model on our list, after Gemma 3 1B, and it should predict faster than the 3B and 3.8B models on the same GPU. It is still about ten times bigger than GPT-2 small, but comfortable on a free T4.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How might it improve our project?&lt;/strong&gt; First, it is a newer model trained on far more data than GPT-2, so we expect better predictions on ordinary English text. Second, and more interesting for us, it was trained on more languages and has a vocabulary of about 150,000 tokens, compared with GPT-2's 50,000. That might mean fewer pieces per Kinyarwanda word. We have not tested this yet, and none of the five models officially lists Kinyarwanda, so we expect imperfect results.&lt;/p&gt;

&lt;h3&gt;
  
  
  How we will check our choice
&lt;/h3&gt;

&lt;p&gt;We will run the same two example sentences from our GPT-2 notebook through Qwen2.5-1.5B and compare three things: tokens per word in English and Kinyarwanda, the loss on each sentence, and the memory and time used on the T4. If Qwen2.5 does not beat GPT-2 on Kinyarwanda, we will say so and keep our tiny GPT, which learned from Kinyarwanda news, for that language.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why not the others?
&lt;/h3&gt;

&lt;p&gt;Phi-3.5-mini is our runner-up for its reasoning, but it is more than twice the size, mostly English, and we have no use for 128K of context. Llama 3.2 3B is gated and has a custom licence, which fails our "no account" rule. Gemma 3 1B is gated too and has the least capacity, although its multilingual vocabulary makes it the next model we would test. SmolLM2-1.7B is free and light, but its short context and English focus make it a weaker fit.&lt;/p&gt;

&lt;h3&gt;
  
  
  Limits
&lt;/h3&gt;

&lt;p&gt;A 1.5B model is still small and will make mistakes, especially in Kinyarwanda, which is thin in its training data. Predicting the next word also means choosing a likely word, not a true one.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. References
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Qwen team. &lt;em&gt;Qwen2.5-1.5B-Instruct&lt;/em&gt; model card. &lt;a href="https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct" rel="noopener noreferrer"&gt;https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Meta. &lt;em&gt;Llama-3.2-3B-Instruct&lt;/em&gt; model card and Llama 3.2 Community License. &lt;a href="https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct" rel="noopener noreferrer"&gt;https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Microsoft. &lt;em&gt;Phi-3.5-mini-instruct&lt;/em&gt; model card. &lt;a href="https://huggingface.co/microsoft/Phi-3.5-mini-instruct" rel="noopener noreferrer"&gt;https://huggingface.co/microsoft/Phi-3.5-mini-instruct&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Google DeepMind. &lt;em&gt;Gemma 3 1B IT&lt;/em&gt; model card and Gemma Terms of Use. &lt;a href="https://huggingface.co/google/gemma-3-1b-it" rel="noopener noreferrer"&gt;https://huggingface.co/google/gemma-3-1b-it&lt;/a&gt; and &lt;a href="https://ai.google.dev/gemma/terms" rel="noopener noreferrer"&gt;https://ai.google.dev/gemma/terms&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Hugging Face. &lt;em&gt;SmolLM2-1.7B-Instruct&lt;/em&gt; model card. &lt;a href="https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct" rel="noopener noreferrer"&gt;https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Radford, A. et al. &lt;em&gt;Language Models are Unsupervised Multitask Learners&lt;/em&gt; (GPT-2), OpenAI, 2019.&lt;/li&gt;
&lt;li&gt;Qwen Team. &lt;em&gt;Qwen2 Technical Report&lt;/em&gt;.&lt;/li&gt;
&lt;li&gt;Google DeepMind. &lt;em&gt;Gemma 3 Technical Report&lt;/em&gt;.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>programming</category>
      <category>beginners</category>
      <category>ai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>what i learn in Introduction to Computer Vision: How Computers See Images my First Computer Vision Journey</title>
      <dc:creator>Grace UMUTONIWASE</dc:creator>
      <pubDate>Tue, 13 Oct 2026 12:43:19 +0000</pubDate>
      <link>https://dev.to/grace_umutoniwase_b6e978d/what-i-learn-in-introduction-to-computer-vision-33pj</link>
      <guid>https://dev.to/grace_umutoniwase_b6e978d/what-i-learn-in-introduction-to-computer-vision-33pj</guid>
      <description>&lt;h1&gt;
  
  
  Introduction to Computer Vision — My Learning Journey
&lt;/h1&gt;

&lt;h2&gt;
  
  
  1. Introduction
&lt;/h2&gt;

&lt;p&gt;Hello everyone, welcome to my blog!&lt;/p&gt;

&lt;p&gt;Today, I want to share my learning journey in Artificial Intelligence, especially Computer Vision.&lt;/p&gt;

&lt;p&gt;In our previous lessons, I learned about Natural Language Processing, which helps computers understand human language. This time, I started learning about Computer Vision, which helps computers understand images and videos.&lt;/p&gt;

&lt;p&gt;During this lesson, I explored how a computer processes an image, how it recognizes objects, and how we can use AI to understand visual information.&lt;/p&gt;

&lt;p&gt;I also tried some practical activities using a computer vision notebook. Let me take you through what I learned, what I tried, and what I discovered.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. What Is Computer Vision?
&lt;/h2&gt;

&lt;p&gt;First, let me explain what Computer Vision means.&lt;/p&gt;

&lt;p&gt;Computer Vision is a field of Artificial Intelligence that helps computers extract meaning from images and videos.&lt;/p&gt;

&lt;p&gt;For example, when I look at a photo, I can easily recognize a person, a car, a cat, or a cup of coffee. But a computer does not see these things the same way I do. It receives numerical information from the image and uses a model to recognize patterns and make predictions.&lt;/p&gt;

&lt;p&gt;The main questions Computer Vision tries to answer are:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;What is in the image?&lt;/li&gt;
&lt;li&gt;Where is it located?&lt;/li&gt;
&lt;li&gt;What is happening in the video?&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  3. How Does a Computer See an Image?
&lt;/h2&gt;

&lt;p&gt;Next, I learned that an image is made up of tiny squares called pixels.&lt;/p&gt;

&lt;p&gt;Each pixel contains numerical values representing its color. For a normal color image, we use three channels: Red, Green, and Blue, also known as RGB.&lt;/p&gt;

&lt;p&gt;For example, a small image with 400 pixels in height and 600 pixels in width contains 720,000 color values because we multiply 400 by 600 by 3.&lt;/p&gt;

&lt;p&gt;This means that a computer starts with numbers, and the challenge is to use those numbers to understand what the image represents.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. What I Learned About Filters
&lt;/h2&gt;

&lt;p&gt;I learned about filters and how they help computers find patterns in images.&lt;/p&gt;

&lt;p&gt;I tried to understand how a filter works by looking at a small grid of numbers called a kernel. The filter moves across the image and performs mathematical calculations using nearby pixels.&lt;/p&gt;

&lt;p&gt;I discovered that different filters have different purposes. For example, a blur filter smooths an image, while a Sobel filter helps detect edges.&lt;/p&gt;

&lt;p&gt;I also learned that the Sobel X filter detects changes in brightness from left to right, which helps identify vertical edges. The Sobel Y filter detects changes from top to bottom, helping identify horizontal edges.&lt;/p&gt;

&lt;p&gt;This showed me that even simple mathematical operations can help a computer find useful visual information.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. What I Learned About Classification, Object Detection, and Segmentation
&lt;/h2&gt;

&lt;p&gt;After learning about pixels and filters, I moved to three important tasks in Computer Vision: classification, object detection, and segmentation.&lt;/p&gt;

&lt;h3&gt;
  
  
  5.1 Image Classification
&lt;/h3&gt;

&lt;p&gt;First, I learned about image classification.&lt;/p&gt;

&lt;p&gt;Image classification means giving one label to an entire image. For example, if I upload a photo of a cat, a classification model may predict that the image contains a cat.&lt;/p&gt;

&lt;p&gt;I learned that the model analyzes the image and predicts which category best matches what it sees.&lt;/p&gt;

&lt;p&gt;I discovered that classification is useful when we want to know what an image contains, but we do not need to know the exact location of each object.&lt;/p&gt;

&lt;h3&gt;
  
  
  5.2 Object Detection
&lt;/h3&gt;

&lt;p&gt;Next, I learned about object detection.&lt;/p&gt;

&lt;p&gt;Object detection helps a computer identify objects in an image and show where they are located.&lt;/p&gt;

&lt;p&gt;I tried to understand how the model uses bounding boxes. A bounding box is a rectangle drawn around a detected object.&lt;/p&gt;

&lt;p&gt;For example, if an image contains a bus and several people, an object detection model may identify the bus and the people and draw boxes around them. It can also display labels and confidence scores for its predictions.&lt;/p&gt;

&lt;p&gt;I discovered that object detection gives us more information than classification because it tells us both what the model detects and where the objects are located.&lt;/p&gt;

&lt;h3&gt;
  
  
  5.3 Image Segmentation
&lt;/h3&gt;

&lt;p&gt;After object detection, I learned about image segmentation.&lt;/p&gt;

&lt;p&gt;Segmentation identifies which pixels belong to an object or a region of an image.&lt;/p&gt;

&lt;p&gt;For example, instead of drawing a rectangle around a person, a segmentation model can mark the pixels that form the person's body.&lt;/p&gt;

&lt;p&gt;I learned that segmentation provides more detailed information about the shape of an object than a simple bounding box.&lt;/p&gt;

&lt;p&gt;I discovered that these three tasks have different purposes: classification gives a label to an image, object detection identifies objects and their locations, and segmentation identifies the pixels belonging to objects.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Trying My Own Photo
&lt;/h2&gt;

&lt;p&gt;After learning about classification, object detection, and segmentation, I decided to test a Computer Vision model using my own example image.&lt;/p&gt;

&lt;h3&gt;
  
  
  My Experiment with a Hot-Air Balloon Photo
&lt;/h3&gt;

&lt;p&gt;For my practical experiment, I used a photo of colorful hot-air balloons floating in the sky at sunset. I uploaded the image to the Computer Vision notebook and ran the object detection model to see what it would recognize.&lt;/p&gt;

&lt;p&gt;The result surprised me. Instead of identifying the objects as hot-air balloons, the model labeled four detections as &lt;strong&gt;"kite."&lt;/strong&gt; The confidence scores displayed were approximately 89%, 91%, 80%, and 83%.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fk88nykkfekt3paoayp6f.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fk88nykkfekt3paoayp6f.png" alt=" " width="800" height="513"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;To make my learning more practical, I tested the object detection model using a photo of colorful hot-air balloons floating in the sky at sunset. I wanted to see whether the model could correctly identify the objects in my image.&lt;/p&gt;

&lt;p&gt;Below is the result I obtained when I ran the model. I was curious to see what labels it would assign to the objects and how confident it would be in its predictions.&lt;/p&gt;

&lt;p&gt;However, the objects in my photo were hot-air balloons, not kites. This means that the model's predictions were incorrect, even though its confidence scores were relatively high.&lt;/p&gt;

&lt;p&gt;This experiment taught me an important lesson: a high confidence score does not necessarily mean that a model is correct. The model can recognize visual patterns but still confuse one object with another.&lt;/p&gt;

&lt;p&gt;I learned why it is important to test Computer Vision models using real images and carefully examine their predictions. Instead of assuming that an AI model is always right, we should compare its results with what is actually present in the image.&lt;/p&gt;

&lt;p&gt;This was one of the most interesting parts of my learning journey because I could see a real example of both the capabilities and the limitations of Computer Vision.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. When Computer Vision Models Fail
&lt;/h2&gt;

&lt;p&gt;One of the most interesting things I learned was that Computer Vision models do not always give correct answers.&lt;/p&gt;

&lt;p&gt;Before this experiment, I might have thought that a trained model could recognize objects correctly in every image. However, I discovered that changing the appearance or quality of an image can affect the model's predictions.&lt;/p&gt;

&lt;h3&gt;
  
  
  7.1 The Original Image
&lt;/h3&gt;

&lt;p&gt;In the original image used in our notebook demonstration, the object detection model detected one bus and four people.&lt;/p&gt;

&lt;p&gt;This gave me an example of how a model can identify several objects in one image and show their locations using bounding boxes.&lt;/p&gt;

&lt;h3&gt;
  
  
  7.2 A Very Dark Image
&lt;/h3&gt;

&lt;p&gt;Next,i tested a very dark version of the image.&lt;/p&gt;

&lt;p&gt;The model made a mistake by detecting a skateboard that was not actually present.&lt;/p&gt;

&lt;p&gt;I learned that poor lighting can make it difficult for a model to recognize objects correctly. It may produce a false detection because the image does not provide clear visual information.&lt;/p&gt;

&lt;h3&gt;
  
  
  7.3 An Upside-Down Image
&lt;/h3&gt;

&lt;p&gt;I also tested an upside-down version of the image.&lt;/p&gt;

&lt;p&gt;In this experiment, the model failed to detect the bus.&lt;/p&gt;

&lt;p&gt;This helped me understand that changing the orientation of an image can affect a model's ability to recognize objects.&lt;/p&gt;

&lt;h3&gt;
  
  
  7.4 A Low-Resolution Image
&lt;/h3&gt;

&lt;p&gt;Finally, we tested a low-resolution version of the image.&lt;/p&gt;

&lt;p&gt;The model did not detect any objects.&lt;/p&gt;

&lt;p&gt;I learned that when an image loses too much detail, the model may struggle to identify what is inside it.&lt;/p&gt;

&lt;h3&gt;
  
  
  7.5 What I Discovered
&lt;/h3&gt;

&lt;p&gt;From these experiments, I discovered that Computer Vision models have limitations.&lt;/p&gt;

&lt;p&gt;Their predictions can be affected by lighting, image orientation, and image resolution.&lt;/p&gt;

&lt;p&gt;I also learned that we should not automatically trust every prediction made by an AI model. We need to examine its results and understand when it can make mistakes.&lt;/p&gt;

&lt;p&gt;This is important because Computer Vision is used in real-world applications such as agriculture, healthcare, security, and environmental conservation. In these areas, incorrect predictions can have serious consequences.&lt;/p&gt;

&lt;h2&gt;
  
  
  8. My Final Conclusion
&lt;/h2&gt;

&lt;p&gt;This is the end of my introduction to Computer Vision.&lt;/p&gt;

&lt;p&gt;During this learning journey, I learned how computers process images, how filters help us detect edges, and how AI models recognize objects using classification, object detection, and segmentation.&lt;/p&gt;

&lt;p&gt;I also discovered that Computer Vision models can make mistakes when images are dark, upside down, or have low resolution.&lt;/p&gt;

&lt;p&gt;My biggest lesson is that Artificial Intelligence is powerful, but we should understand its limitations and test its predictions carefully.&lt;/p&gt;

&lt;p&gt;I enjoyed learning these concepts through practical experiments because they helped me connect theory with real-world applications.&lt;/p&gt;

&lt;p&gt;This is only the beginning of my journey in Artificial Intelligence. I look forward to learning more, building projects, and discovering how Computer Vision can solve real-world problems, especially in Africa.&lt;/p&gt;

&lt;p&gt;Thank you for reading my vlog and following my learning journey. &lt;/p&gt;

&lt;p&gt;See you in my next blog!&lt;/p&gt;

</description>
      <category>ai</category>
      <category>programming</category>
      <category>python</category>
      <category>architecture</category>
    </item>
    <item>
      <title>My First Steps into Word Embeddings with Word2Vec</title>
      <dc:creator>Grace UMUTONIWASE</dc:creator>
      <pubDate>Thu, 01 Oct 2026 15:07:26 +0000</pubDate>
      <link>https://dev.to/grace_umutoniwase_b6e978d/my-first-steps-into-word-embeddings-with-word2vec-13l1</link>
      <guid>https://dev.to/grace_umutoniwase_b6e978d/my-first-steps-into-word-embeddings-with-word2vec-13l1</guid>
      <description>&lt;ol&gt;
&lt;li&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Explain what word embeddings are and why they matter: so, Word embeddings are numerical representations of words. Each word is represented by a list of numbers called a vector.&lt;br&gt;
For example, imagine representing two words using simplified vectors:&lt;/p&gt;

&lt;p&gt;happy = [0.8, 0.7, 0.2]&lt;br&gt;
joyful = [0.7, 0.8, 0.3]&lt;/p&gt;

&lt;p&gt;These numbers are only illustrative, not real trained embeddings.&lt;/p&gt;

&lt;p&gt;And why they matter :Word embeddings are useful because many NLP tasks require a computer to process language beyond simply identifying individual words.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Cover at least one embedding method which is Word2Vec.so,Word2Vec is a popular technique used in Natural Language Processing (NLP) to learn word embeddings. It converts words into numerical vectors by learning from the contexts in which words appear in a large collection of text.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Word2Vec is a method that teaches a computer to understand relationships between words by learning from the words that appear around them.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;How does it work?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;. It reads sentences&lt;/p&gt;

&lt;p&gt;For example: “I love learning Python.”&lt;/p&gt;

&lt;p&gt;Word2Vec looks at the words and how they appear together.&lt;br&gt;
How does it work?&lt;/p&gt;

&lt;p&gt;. It learns from context&lt;/p&gt;

&lt;p&gt;If it sees “I love learning Python” and “I enjoy learning Python,” it notices that love and enjoy appear in similar contexts.&lt;/p&gt;

&lt;p&gt;. It converts words into numbers&lt;/p&gt;

&lt;p&gt;During training, Word2Vec learns a numerical vector for each word. Words used in similar contexts may end up with similar vectors.&lt;/p&gt;

&lt;p&gt;. It learns relationships&lt;/p&gt;

&lt;p&gt;After training, we can use the vectors to find words that are similar in context.&lt;/p&gt;

&lt;p&gt;Two ways Word2Vec learns&lt;/p&gt;

&lt;p&gt;CBOW: Uses surrounding words to predict a missing word.&lt;/p&gt;

&lt;p&gt;Skip-gram: Uses one word to predict the surrounding words.&lt;/p&gt;

&lt;p&gt;Sentences → Context → Training → Word vectors → Word relationships`&lt;/p&gt;

&lt;p&gt;Word2Vec does not understand language exactly like a human. It learns patterns from the text it receives. If its training data is too small, its results may not be reliable.&lt;/p&gt;

&lt;p&gt;. Include at least one concrete example: a code snippet, a similarity result, an analogy, or a&lt;br&gt;
visualisation.&lt;/p&gt;

&lt;p&gt;&amp;lt;&amp;gt;&lt;br&gt;
from gensim.models import Word2Vec&lt;/p&gt;

&lt;h1&gt;
  
  
  Example training sentences
&lt;/h1&gt;

&lt;p&gt;sentences = [&lt;br&gt;
    ["i", "love", "learning", "python"],&lt;br&gt;
    ["i", "enjoy", "learning", "python"],&lt;br&gt;
    ["i", "love", "learning", "mathematics"],&lt;br&gt;
    ["python", "is", "interesting"],&lt;br&gt;
    ["mathematics", "is", "interesting"]&lt;br&gt;
]&lt;/p&gt;

&lt;p&gt;4.Share something you learned, found surprising, or struggled with :&lt;/p&gt;

&lt;p&gt;What I Learned and Found Surprising&lt;/p&gt;

&lt;p&gt;One thing I learned is that computers can learn relationships between words by studying the words that appear around them. I found it surprising that Word2Vec can convert words into numerical vectors and use those vectors to identify words that are used in similar contexts.&lt;/p&gt;

&lt;p&gt;At first, I struggled to understand how words could be represented by numbers and how those numbers could show relationships between words. After learning about CBOW and Skip-gram, I understood that Word2Vec learns from context: CBOW predicts a word from its surrounding words, while Skip-gram predicts surrounding words from a given word.&lt;/p&gt;

&lt;p&gt;This experience helped me understand how Natural Language Processing (NLP) allows computers to work with human language. I also learned that the quality of the results depends on the amount and quality of the training data.&lt;/p&gt;

&lt;p&gt;Learning about Word2Vec has helped me understand an important concept in Natural Language Processing. I now understand how a model can learn word vectors by studying surrounding words and how these vectors can be used to explore similarities between words.&lt;/p&gt;

</description>
      <category>nlp</category>
      <category>python</category>
      <category>machinelearning</category>
    </item>
  </channel>
</rss>
