NVIDIA is a multinational company known for its computing hardware, especially its graphics processing units (GPUs) and systems on chip units (SoCs) for mobile devices. The company went public on January 22, 1999.
While the company continues to focus on hardware manufacturing, it has implemented deep learning and AI in its GPUs and specific software, such as the autonomous driving platform.
The company trades on the NASDAQ (symbol: NVDA) with a Market capitalization of just over $418 billion and employs approximately 23,000 worldwide. in his annual reportthe company reports approximately $27 billion in revenue for the fiscal year ending 2021.
In this article, we’ll explore how NVIDIA has implemented AI applications for its business and industry through two discrete use cases:
- Improved video resolution – Nvidia uses machine learning to improve video resolution and audio quality during real-time video calls.
- Analysis of the impact of marketing campaigns on sales – NVIDIA uses ML algorithms and data analytics to analyze marketing touchpoints and optimize marketing campaigns for a new range of GPUs.
Use Case #1 – Using Machine Learning to Improve Video Quality
NVIDIA claims its research scientists encountered a problem that prevented virtually everyone who works from home: audio and video interruptions and video call crashes. The company claims that these video conferencing issues disrupted critical work processes.
Streaming video calls requires a lot of bandwidth, often resulting in non-ideal communications: frozen video and/or audio, poor image resolution, broken audio, and so on.
On the video side, one thing that increases bandwidth requirements is a software called video codec, which most organizations use to compress and decompress video. On the audio side, external background noise can be disruptive. Let’s take a look at what NVIDIA did to address both.
NVIDIA claims that traditional video transmission compression software, such as video codecs, transmit a large number of pixels, which requires a lot of data and bandwidth to send and receive.
Instead of using this traditional software, NVIDIA claims it uses a specialized ML model in its Maxine product, a generative contrarian network, or GAN† The company claims that this reduces bandwidth ten times over conventional video transmission methods.
The company claims that its software works by first reducing the number of pixels sent. The model achieves this by extracting user image data limited to key points of view, specifically the user’s eyes, nose, and mouth, rather than the entire image.
The ML model then uses the critical face area data and produces an output in the form of a reconstruction of the original image and subsequent images. NVIDIA claims that much less data is sent over the network with this ML model, resulting in better image quality.
In the following video of about 2.5 minutes, the company explains how its model works:
In addition to improving image resolution, the company claims the ML model also reduces camera noise in low light and applies “AI-powered background removal, replacement or blurring. The company also claims it works regardless of whether the user is wearing a mask or hat.
On the audio side, NVIDIA states that it uses AI to both filter out unwanted background noise (i.e. noise reduction) and improve audio quality, especially speech. However, the company doesn’t reveal how it does this, other than that the AI uses what appears to be neural networks to improve the quality of the user’s speech.
As for tangible business results, our research failed to uncover any.
Use Case #2 – Predicting customer churn
NVIDIA wanted to determine how its marketing strategy affected sales of a new GPU line. Adobe has determined that measuring NVIDIAs marketing effectiveness was the best way to achieve the client’s goals. With this data, NVIDIA could potentially see the financial contribution of each marketing channel and, if desired, adjust its marketing campaign to achieve the desired business outcomes.
It appears that Adobe’s method of quantifying NVIDIA’s marketing campaigns involved analyzing multiple data sets to identify and analyze the most influential channels for acquiring customers, increasing conversions, and driving incremental sales. Adobe states that it uses software called Attribution AI for this.
Adobe first acquired the relevant market and sales data sets. The marketing and sales data was fed into Adobe’s Attribution AI platform, which used an AI algorithm based on a statistical technique called cooperative game theory† Adobe claims that it uses this algorithm because marketing touchpoints are “often misrepresented and undervalued or overvalued” with traditional rules-based models.
Below is a video of about 3 minutes explaining how the software works:
It is claimed that the output of the algorithm shows the influence and incremental sales for each marketing for each channel through a graph.
In the case of NVIDIA, the GPU marketing campaigns channels (paid media vs email) and campaigns (bundled offers versus brand renewal) were the most influential in driving product sales. It seems that the software also gives NVIDIA the ability to analyze the incremental sales of each marketing channel individually and in aggregate.
It seems that NVIDIA can select individual channels such as paid media, organic social media, billboards, email, affiliate and third-party channels and see their respective influence on both marketing and incremental sales.
Regarding tangible business outcomes, Adobe claims its solution gave NVIDIA the ability to quantify incremental sales and marketing effectiveness across channels. More broadly, Adobe claims its software-enabled NVIDIA can get a more accurate picture of its marketing effectiveness by isolating last touch attributionwhich may skew the analysis.
Specifically, no quantitative evidence is provided to support Adobe’s claims of successful business results for NVIDIA.