added short captions for the videos

contents/ai_for_good/ai_for_good.qmd

@@ -164,10 +164,14 @@ In disaster response, speed and safety are paramount. But rubble and wreckage cr
 
 When buildings collapse after earthquakes, small drones can prove invaluable. Equipped with TinyML navigation algorithms, micro-sized drones like the [CrazyFlie](https://www.bitcraze.io/) can traverse cramped voids and map pathways beyond human reach [@duisterhof2019learning]. Obstacle avoidance allows the drones to weave through unstable debris. This autonomous mobility lets them rapidly sweep areas humans cannot access.
 
+The video below is a presentation of the [@duisterhof2019learning] paper on deep reinforcement learning for source seeking using drones.
+
 {{< video https://www.youtube.com/watch?v=wmVKbX7MOnU >}}
 
 Crucially, onboard sensors and TinyML processors analyze real-time data to identify signs of survivors. Thermal cameras detect body heat, microphones pick up calls for help, and gas sensors warn of leaks [@duisterhof2021sniffy]. Processing data locally using TinyML allows for quick interpretation to guide rescue efforts. As conditions evolve, the drones can adapt by adjusting their search patterns and priorities.
 
+The following video is an overview of autonomous drones for gas leak detection.
+
 {{< video https://www.youtube.com/watch?v=hj_SBSpK5qg >}}
 
 Additionally, coordinated swarms of drones unlock new capabilities. By collaborating and sharing insights, drone teams achieve a comprehensive view of the situation. Blanketing disaster sites allows TinyML algorithms to fuse and analyze data from multiple vantage points. This amplifies situational awareness beyond individual drones [@duisterhof2021sniffy].
@@ -196,6 +200,8 @@ With machine learning algorithms running locally on microcontrollers, compact ac
 
 Similarly, mobility devices could use on-device vision processing to identify obstacles and terrain characteristics. This enables enhanced navigation and safety for the visually impaired. Companies like [Envision](https://www.letsenvision.com/) are developing smart glasses, converting visual information into speech, with embedded TinyML to guide the blind by detecting objects, text, and traffic signals.
 
+The video below shows the different real life usecases of the Envision visual aid glasses.
+
 {{< video https://www.youtube.com/watch?v=oGWinIKDOdc >}}
 
 TinyML could even power responsive prosthetic limbs. By analyzing nerve signals and sensory data like muscle tension, prosthetics and exoskeletons with embedded ML can move and adjust grip dynamically. This makes control more natural and intuitive. Companies are creating affordable, everyday bionic hands using TinyML. And for those with speech difficulties, voice-enabled devices with TinyML can generate personalized vocal outputs from non-verbal inputs. Pairs by Anthropic translates gestures into natural speech tailored for individual users.

contents/dl_primer/dl_primer.qmd

@@ -91,6 +91,8 @@ Multi-layer perceptrons (MLPs) are an evolution of the single-layer perceptron m
 
 The forward pass is the initial phase where data moves through the network from the input to the output layer. During this phase, each layer performs specific computations on the input data, using weights and biases before passing the resulting values to subsequent layers. The final output of this phase is used to compute the loss, indicating the difference between the predicted output and actual target values.
 
+The video below explains how neural networks work using handwritten digits recognition as an example application. It also touches on the math underlying neural nets.
+
 {{< video https://www.youtube.com/embed/aircAruvnKk?si=qfkBf8MJjC2WSyw3 >}}
 
 #### Backward Pass (Backpropagation)
@@ -99,6 +101,7 @@ Backpropagation is a key algorithm in training deep neural networks. This phase
 
 Grasping these foundational concepts paves the way to understanding more intricate deep learning architectures and techniques, fostering the development of more sophisticated and efficacious applications, especially within the realm of embedded AI systems.
 
+The following two videos build upon the previous one. They cover gradient descent and backprogation in neural networks.
 
 {{< video https://www.youtube.com/watch?v=IHZwWFHWa-w&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=2 >}}
 

contents/hw_acceleration/hw_acceleration.qmd

@@ -765,7 +765,9 @@ The wafer-scale approach also diverges from more modular system-on-chip designs
 
 By designing the wafer as one integrated logic unit, data transfer between elements is minimized. This provides lower latency and power consumption compared to discrete system-on-chip or chiplet designs. While chiplets can offer flexibility by mixing and matching components, communication between chiplets is a challenge. The monolithic nature of wafer-scale integration eliminates these inter-chip communication bottlenecks.
 
-However, the ultra-large scale also poses difficulties for manufacturability and yield with wafer-scale designs. Defects in any region of the wafer can make (certian parts of) the chip unusable. And specialized lithography techniques are required to produce such large dies. So wafer-scale integration pursues the maximum performance gains from integration but requires overcoming substantial fabrication challenges. The following video will provide additional context.
+However, the ultra-large scale also poses difficulties for manufacturability and yield with wafer-scale designs. Defects in any region of the wafer can make (certian parts of) the chip unusable. And specialized lithography techniques are required to produce such large dies. So wafer-scale integration pursues the maximum performance gains from integration but requires overcoming substantial fabrication challenges.
+
+The following video will provide additional context.
 
 {{< video https://www.youtube.com/watch?v=Fcob512SJz0 >}}
 
@@ -800,7 +802,9 @@ Neuromorphic computing is an emerging field aiming to emulate the efficiency and
 
 Intel and IBM are leading commercial efforts in neuromorphic hardware. Intel's Loihi and Loihi 2 chips [@davies2018loihi; @davies2021advancing] offer programmable neuromorphic cores with on-chip learning. IBM's Northpole [@modha2023neural] device comprises more than 100 million magnetic tunnel junction synapses and 68 billion transistors. These specialized chips deliver benefits like low power consumption for edge inference.
 
-Spiking neural networks (SNNs) [@maass1997networks] are computational models suited for neuromorphic hardware. Unlike deep neural networks that communicate via continuous values, SNNs use discrete spikes more akin to biological neurons. This allows efficient event-based computation rather than constant processing. Additionally, SNNs take into account temporal characteristics of input data in addition to spatial characteristics. This better mimics biological neural networks, where timing of neuronal spikes plays an important role. However, training SNNs remains challenging due to the added temporal complexity. @fig-spiking provides an overview of the spiking methodlogy: (a) Diagram of a neuron; (b) Measuring an action potential propagated along the axon of a neuron. Only the action potential is detectable along the axon; (c) The neuron's spike is approximated with a binary representation; (d) Event-Driven Processing; (e) Active Pixel Sensor and Dynamic Vision Sensor. You can also watch the video linked below for a more detailed explanation.
+Spiking neural networks (SNNs) [@maass1997networks] are computational models suited for neuromorphic hardware. Unlike deep neural networks that communicate via continuous values, SNNs use discrete spikes more akin to biological neurons. This allows efficient event-based computation rather than constant processing. Additionally, SNNs take into account temporal characteristics of input data in addition to spatial characteristics. This better mimics biological neural networks, where timing of neuronal spikes plays an important role. However, training SNNs remains challenging due to the added temporal complexity. @fig-spiking provides an overview of the spiking methodlogy: (a) Diagram of a neuron; (b) Measuring an action potential propagated along the axon of a neuron. Only the action potential is detectable along the axon; (c) The neuron's spike is approximated with a binary representation; (d) Event-Driven Processing; (e) Active Pixel Sensor and Dynamic Vision Sensor.
+
+You can also watch the video linked below for a more detailed explanation.
 
 ![Neuromoprhic spiking. Credit: @eshraghian2023training.](images/png/aimage4.png){#fig-spiking}
 
@@ -990,7 +994,9 @@ Realizing foundation models for end-to-end hardware design will require:
 
 While significant research remains, foundation models represent the most transformative long-term goal for imbuing AI into the hardware design process. Democratizing hardware design via versatile, automated ML systems promises to unlock a new era of optimized, efficient, and innovative chip design. The journey ahead is filled with open challenges and opportunities.
 
-We encourage you to read [Architecture 2.0](https://www.sigarch.org/architecture-2-0-why-computer-architects-need-a-data-centric-ai-gymnasium/) if ML-aided computer architecture design [@krishnan2023archgym] interests you. Alternatively, you can watch the below video.
+We encourage you to read [Architecture 2.0](https://www.sigarch.org/architecture-2-0-why-computer-architects-need-a-data-centric-ai-gymnasium/) if ML-aided computer architecture design [@krishnan2023archgym] interests you.
+
+Alternatively, you can watch the below video.
 
 {{< video https://www.youtube.com/watch?v=F5Eieaz7u1I&ab_channel=OpenComputeProject >}}
 

contents/ondevice_learning/ondevice_learning.qmd

@@ -394,7 +394,7 @@ The complexity of the source model can also impact the success of transfer learn
 
 By considering these factors, ML practitioners can make informed decisions about when and how to utilize transfer learning, ultimately leading to more successful model performance on the target task. The success of transfer learning hinges on the degree of similarity between the source and target domains. There is risk of overfitting, especially when fine-tuning occurs on a limited dataset. On the computational front, it is worth noting that certain pre-trained models, owing to their size, might not comfortably fit into the memory constraints of some devices or may run prohibitively slowly. Over time, as data evolves, there is potential for model drift, indicating the need for periodic re-training or ongoing adaptation.
 
-Learn more about transfer learning in the video below:
+Learn more about transfer learning in the video below.
 
 {{< video https://www.youtube.com/watch?v=FQM13HkEfBk >}}
 
@@ -416,7 +416,7 @@ This boils down the basic algorithm for federated learning on the right. For eac
 
 With this proposed structure, there are a few key vectors for optimizing federated learning further. We will outline each in the following subsections.
 
-The following video is an overview of federated learning:
+The following video is an overview of federated learning.
 
 {{< video https://www.youtube.com/watch?v=zqv1eELa7fs >}}
 

contents/ops/ops.qmd

@@ -88,7 +88,7 @@ Here is a table that summarizes them side by side.
 | **Primary Concerns** | Code integration, Testing, Release management, Automation, Infrastructure as code | Data management, Model versioning, Experiment tracking, Model deployment, Scalability of ML workflows |
 | **Typical Outcomes** | Faster and more reliable software releases, Improved collaboration between development and operations teams | Efficient management and deployment of machine learning models, Enhanced collaboration between data scientists and engineers |
 
-Learn more about ML Lifecycles through a case study featuring speech recognition:
+Learn more about ML Lifecycles through a case study featuring speech recognition.
 
 {{< video https://www.youtube.com/watch?v=YJsRD_hU4tc&list=PLkDaE6sCZn6GMoA0wbpJLi3t34Gd8l0aK&index=3 >}}
 
@@ -112,7 +112,7 @@ For instance, a pipeline may ingest data from [PostgreSQL](https://www.postgresq
 
 In an industrial predictive maintenance use case, sensor data is ingested from devices into S3. A Prefect pipeline processes the sensor data, joining it with maintenance records. The enriched dataset is stored in Feast so models can easily retrieve the latest data for training and predictions.
 
-Here is a short overview of data pipelines:
+The video below is a short overview of data pipelines.
 
 {{< video https://www.youtube.com/watch?v=gz-44N3MMOA&list=PLkDaE6sCZn6GMoA0wbpJLi3t34Gd8l0aK&index=33 >}}
 
@@ -214,7 +214,7 @@ Teams configure alerting for key monitoring metrics like accuracy declines and s
 
 Comprehensive monitoring enables teams to maintain confidence in model and system health after deployment. It empowers teams to catch and resolve deviations through data-driven alerts and dashboards preemptively. Active monitoring is essential for maintaining highly available, trustworthy ML systems.
 
-Watch the video below to learn more about monitoring:
+Watch the video below to learn more about monitoring.
 
 {{< video https://www.youtube.com/watch?v=hq_XyP9y0xg&list=PLkDaE6sCZn6GMoA0wbpJLi3t34Gd8l0aK&index=7 >}}
 
@@ -242,7 +242,7 @@ For example, a data scientist may use Weights & Biases to analyze an anomaly det
 
 Enabling transparency, traceability and communication via MLOps empowers teams to remove bottlenecks and accelerate delivery of impactful ML systems.
 
-The following video covers key challenges in model deployment, including concept drift, model drift, and software engineering issues:
+The following video covers key challenges in model deployment, including concept drift, model drift, and software engineering issues.
 
 {{< video https://www.youtube.com/watch?v=UyEtTyeahus&list=PLkDaE6sCZn6GMoA0wbpJLi3t34Gd8l0aK&index=5 >}}
 

contents/optimizations/optimizations.qmd

@@ -707,6 +707,8 @@ In addition to being an indispensable technique for many edge processors, quanti
 
 Thus, quantization combined with efficient low-precision logic and dedicated deep learning accelerators, has been one crucial driving force for the evolution of such edge processors.
 
+The video below is a lecture on quantization and the different quantization methods.
+
 {{< video https://www.youtube.com/watch?v=AlASZb93rrc >}}
 
 ## Efficient Hardware Implementation {#sec-model_ops_hw}

contents/privacy_security/privacy_security.qmd

@@ -86,6 +86,8 @@ This breach was significant due to its sophistication; Stuxnet specifically targ
 
 The Jeep Cherokee hack was a groundbreaking event demonstrating the risks inherent in increasingly connected automobiles [@miller2019lessons]. In a controlled demonstration, security researchers remotely exploited a vulnerability in the Uconnect entertainment system, which had a cellular connection to the internet. They were able to control the vehicle's engine, transmission, and brakes, alarming the automotive industry into recognizing the severe safety implications of cyber vulnerabilities in vehicles.
 
+The video below is a short documentary of the attack.
+
 {{< video https://www.youtube.com/watch?v=MK0SrxBC1xs&ab_channel=WIRED title="Hackers Remotely Kill a Jeep on a Highway" >}}
 
 While this wasn't an attack on an ML system per se, the reliance of modern vehicles on embedded systems for safety-critical functions has significant parallels to the deployment of ML in embedded systems, underscoring the need for robust security at the hardware level.
@@ -94,6 +96,8 @@ While this wasn't an attack on an ML system per se, the reliance of modern vehic
 
 The Mirai botnet involved the infection of networked devices such as digital cameras and DVR players [@antonakakis2017understanding]. In October 2016, the botnet was used to conduct one of the largest [DDoS](https://www.cloudflare.com/learning/ddos/what-is-a-ddos-attack/) attacks ever, disrupting internet access across the United States. The attack was possible because many devices used default usernames and passwords, which were easily exploited by the Mirai malware to control the devices.
 
+The following video presentation explains how the Mirai Botnet works.
+
 {{< video https://www.youtube.com/watch?v=1pywzRTJDaY >}}
 
 Although the devices were not ML-based, the incident is a stark reminder of what can happen when numerous embedded devices with poor security controls are networked together, a situation that is becoming more common with the growth of ML-based IoT devices.
@@ -356,7 +360,7 @@ Now, let's look at the power analysis chart when we enter an incorrect password
 
 The example above shows how we can infer information about the encryption process and the secret key itself through analyzing different inputs and try to 'eavesdrop' on the operations that the device is performing on each byte of the input.
 
-For additional details, please see the following video:
+For a more detailed explanation, watch the video below.
 
 {{< video <https://www.youtube.com/watch?v=2iDLfuEBcs8> title="ECED4406 - 0x501 Power Analysis Attacks" >}}
 

contents/responsible_ai/responsible_ai.qmd

@@ -86,6 +86,8 @@ Putting AI systems into real-world operation requires ensuring they are safe, re
 
 Adversarial attacks that subtly alter input data can also fool ML models and cause dangerous failures if systems are not resistant. Deepfakes represent another emerging threat area.
 
+Below is a deepfake video of Barack Obama that went viral a few years ago.
+
 {{< video https://www.youtube.com/watch?v=AmUC4m6w1wo&ab_channel=BBCNews >}}
 
 Promoting safety requires extensive testing, risk analysis, human oversight, and designing systems that combine multiple weak models to avoid single points of failure. Rigorous safety mechanisms are essential for the responsible deployment of capable AI.