最新

Hi everyone. I have been learning about deep learning for some time, and I&#x27;ve tried to implement CNN, neural networks, U-Net, transformers etc. to learn and understand them more and also to get my hands dirty on the frameworks, however I&#x27;ve noticed that many tutorials online are not very detailed, so concepts are not explained clearly, so people would understand neural networks only shallowly. On the other hand, many sources like books may show many, many equations but do not show the main points, so people reading those books would get lost in mathematical details, which hampers learning. When I tried reading about RNN, or LSTM, I&#x27;ve noticed that many tutorials do not fully explain them. Some show pictures to make visualization easier, some show forward equations but the backward equations are not discussed. But there is something which I don&#x27;t think is talked about much, is that many tutorials, even if they show the backpropagation, only limit it to a single RNN layer (this is also true for LSTM&#x2F;GRU).<p>Some time ago, I read this book called &quot;Neural Network design&quot; by M. Hagan, and I found the explanations of the book to be quite good (even though the book is not new). The book explains things clearly enough for you to build everything and does not handwaive anything. When I checked the part about RNN, I noticed that the book explains how to do backprop for RNN with arbitrary connections, not just one RNN layer, which I think is something not many sources online show. The book also derives for the conditions of different delays, which I think is completely skipped in other sources.<p>So I decided to go ahead and implement it. The URL provides link to my implementation, which includes: The implementation includes: - Full BPTT for RNN networks with arbitrary recurrent connections and delays - Comprehensive gradient checking using finite differences - Bayesian regularization and multiple optimization algorithms - Extensive numerical validation throughout<p>I think I learned a lot during the implementation, both about how to implement a neural network, and also about how to structure my program, etc. I tried to be systematic and included tests for correctness of backprop by approximate difference equation (you know the [f(x+delta)-f(x-delta)]&#x2F;(2*delta) thing). This also made me try to learn about Einstein summation (using Numpy) which really help things. During this period, I also learned that equation (14.39) has some slight error which is fixed in later equations (this was confirmed in private emails with the authors). The gradient checking was essential for debugging these subtle mathematical issues.<p>Key lessons: - Systematic software development techniques, coupled with mathematical rigour, help catch ML bugs more effectively. - Implementing from first principles help solidify your understanding and reveal the inner workings which frameworks hide. - Einstein summation is a good thing to make the maths much cleaner.<p>I know that this network is quite old, but I just wanted to share with you my experience. Overall I think there is value in firstly grounding in fundamentals before jumping to more complex models.

Built an iOS app that runs a local OCR server using Apple&#x27;s Vision Framework.<p>Creates a REST API endpoint accessible from any device on your network. No cloud services needed - everything processes locally on the phone.<p>Available on App Store (searching &quot;OCR Server&quot;).<p>Would appreciate feedback on the architecture or similar mobile-as-server projects you&#x27;ve seen.

A motion-graphic comparison website in the vein of LMArena. The videos are rendered via Remotion.<p>We hope that AI will be used in interesting ways to help with video production, so we wanted to give some of the models available today a shot at some basic graphics.

Hello HN! Recently, we have released Nocturne 3.0.0, which is a complete replacement for the (now unusable) Spotify Car Thing stock firmware. We&#x27;re proud to eliminate more e-waste in the world.<p># Changes from v2 - Bluetooth tethering for car use (no more Raspberry Pi in the car) - Full graphics acceleration - Native Spotify login (no more client ID&#x2F;secret) - Start DJ from the Car Thing - Podcast support - Gesture control - New settings - Boot to Now Playing - Spotify Connect device switcher - Support for Japanese, Simplified Chinese, Traditional Chinese, Korean, Arabic, Devanagari, Hebrew, Bengali, Tamil, Thai, Cyrillic, Vietnamese, and Greek - Full knob control support - Local file support - Preset button support - Status bar on home (shows time &amp; Bluetooth&#x2F;Wi-Fi) - Auto brightness - Hold settings button for power menu - Lock screen showing time full screen (press settings button) - DJ preset binding (hold preset button while DJ is playing in Now Playing) - Spotify mixes in Radio tab (Discover Weekly, daily mixes, etc.) - OTA updates - + MUCH more (this is just the important stuff!)<p># Flashing A guide to flashing Nocturne 3.0.0 is in the README. Bluetooth will work out of the box, or choose an alternative in the Setting up Network section. Hotspot capability from your phone and plan are required for Bluetooth.<p># Notes This wouldn’t be possible without our donors and the rest of the Nocturne Team. We hope you’ll enjoy it, as we&#x27;ve spent thousands of hours working on it!<p>Consider buying the team a coffee if you can <a href="https:&#x2F;&#x2F;usenocturne.com&#x2F;support" rel="nofollow">https:&#x2F;&#x2F;usenocturne.com&#x2F;support</a><p><a href="https:&#x2F;&#x2F;github.com&#x2F;usenocturne&#x2F;nocturne&#x2F;releases&#x2F;tag&#x2F;v3.0.0" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;usenocturne&#x2F;nocturne&#x2F;releases&#x2F;tag&#x2F;v3.0.0</a>