Neural encoding enables more-efficient recovery of lost audio packets

By leveraging neural vocoding, Amazon Chime SDK’s new deep-redundancy (DRED) technology can reconstruct long sequences of lost packets with little bandwidth overhead.

Packet loss is a big problem for real-time voice communication over the Internet. Everyone has been in the situation where the network is becoming unreliable and enough packets are getting lost that it's hard — or impossible — to make out what the other person is saying.

One way to fight packet loss is through redundancy, in which each new packet includes information about prior packets. But existing redundancy schemes either have limited scope — carrying information only about the immediately preceding packet, for instance — or scale inefficiently.

The Deep REDundancy (DRED) technology from the Amazon Chime SDK team significantly improves quality and intelligibility under packet loss by efficiently transmitting large amounts of redundant information. Our approach leverages the ability of neural vocoders to reconstruct informationally rich speech signals from informationally sparse frequency spectrum snapshots, and we use a neural encoder to compress those snapshots still further. With this approach, we are able to load a single packet with information about as many as 50 prior packets (one second of speech) with minimal increase in bandwidth.

We describe our approach in a paper that we will present at this year’s ICASSP.

Redundant audio

All modern codecs (coder/decoders) have so-called packet-loss-concealment (PLC) algorithms that attempt to guess the content of lost packets. Those algorithms work fine for infrequent, short losses, as they can extrapolate phonemes to fill in gaps of a few tens of milliseconds. However, they cannot (and certainly should not try to) predict the next phoneme or word from the conversation. To deal with significantly degraded networks, we need more than just PLC.

Related content
Combining classic signal processing with deep learning makes method efficient enough to run on a phone.

One option is the 25-year-old spec for REDundant audio data (often referred to as just RED). Despite its age, RED is still in use today and is one of the few ways of transmitting redundant data for WebRTC, a popular open-source framework for real-time communication over the Web. RED has the advantage of being flexible and simple to use, but it is not very efficient. Transmitting two copies of the audio requires ... twice the bitrate.

The Opus audio codec — which is the default codec for WebRTC — introduced a more efficient scheme for redundancy called low-bit-rate redundancy (LBRR). With LBRR, each new audio packet can include a copy of the previous packet, encoded at a lower bit rate. That has the advantage of lowering the bit rate overhead. Also, because the scheme is deeply integrated into Opus, it can be simpler to use than RED.

That being said, the Opus LBRR is limited to just one frame of redundancy, so it cannot do much in the case of a long burst of lost packets. RED does not have that limitation, but transmitting a large number of copies would be impractical due to the overhead. There is always the risk that the extra redundancy will end up causing congestion and more losses.

LBRR and PLC.png
With every new voice packet (blue), Opus’s low-bit-rate-redundancy (LBRR) mechanism includes a compressed copy of the previous packet (green). When three consecutive packets are lost (red x’s), two of them are unrecoverable, and a packet-loss-concealment (PLC) algorithm must fill in the gaps.

Deep REDundancy (DRED)

In the past few years, we have seen neural speech codecs that can produce good quality speech at only a fraction of the bit rate required by traditional speech codecs — typically less than three kilobits per second (3 kb/s). That was unthinkable just a few years ago. But for most real-time-communication applications, neural codecs aren't that useful, because just the packet headers required by the IP/UDP/RTP protocols take up 16 kb/s.

However, for the purpose of transmitting a large amount of redundancy, a neural speech codec can be very useful, and we propose a Deep REDundancy codec that has been specifically designed for that purpose. It has a different set of constraints than a regular speech codec:

  • The redundancy in different packets needs to be independent (that's why we call it redundancy in the first place). However, within each packet, we can use as much prediction and other redundancy elimination as we like since IP packets are all-or-nothing (no corrupted packets).
  • We want to encode meaningful acoustic features rather than abstract (latent) ones to avoid having to standardize more than needed and to leave room for future technology improvements.
  • There is a large degree of overlap between consecutive redundancy packets. The encoder should leverage this overlap and should not need to encode each redundancy packet from scratch. The encoding complexity should remain constant even as we increase the amount of redundancy.
  • Since short bursts are more common than long ones, the redundancy decoder should be able to decode the most recent audio quickly but may take longer to decode older signals.
  • The Opus decoder has to be able to switch between decoding DRED, PLC, LBRR, and regular packets at any time.

Neural vocoders

Let's take a brief detour and discuss neural vocoders. A vocoder is an algorithm that takes in acoustic features that describe the spectrum of a speech signal over a short span of time and generates the corresponding (continuous) speech signal. Vocoders can be used in text-to-speech, where acoustic features are generated from text, and for speech compression, where the encoder transmits acoustic features, and a vocoder generates speech from the features.

Related content
A text-to-speech system, which converts written text into synthesized speech, is what allows Alexa to respond verbally to requests or commands...

Vocoders have been around since the ’70s, but none had ever achieved acceptable speech quality — until neural vocoders like WaveNet came about and changed everything. WaveNet itself was all but impossible to implement in real time (even on a GPU), but it led to lower-complexity neural vocoders, like the LPCNet vocoder we're using here.

Like many (but not all) neural vocoders, LPCNet is autoregressive, in that it produces the audio samples that best fit the previous samples — whether the previous samples are real speech or speech synthesized by LPCNet itself. As we will see below, that property can be very useful.

DRED architecture

The vocoder’s inputs — the acoustic features — don't describe the full speech waveform, but they do describe how the speech sounds to the human ear. That makes them lightweight and predictable and thus ideal for transmitting large amounts of redundancy.

The idea behind DRED is to compress the features as much as possible while ensuring that the recovered speech is still intelligible. When multiple packets go missing, we wait for the first packet to arrive and decode the features it contains. We then send those features to a vocoder — in our case, LPCNet — which re-synthesizes the missing speech for us from the point where the loss occurred. Once the "hole" is filled, we resume with Opus decoding as usual.

Combining the constraints listed earlier leads to the encoder architecture depicted below, which enables efficient encoding of highly redundant acoustic features — so that extended holes can be filled at the decoder.

Codec.png
Every 20 milliseconds, the DRED encoder encodes the last 40 milliseconds of speech. The decoder works backward, as the most recently transmitted audio is usually the most important.

The DRED encoder works as follows. Every 20 milliseconds (ms), it produces a new vector that contains information about the last 40 ms of speech. Given this overlap, we need only half of the vectors to reconstruct the complete speech. To avoid our redundancy’s being itself redundant, in a given 20 ms packet, we include only every other redundancy coding vector, so the redundancy encoded in a given packet covers nonoverlapping segments of the past speech. In terms of the figure above, the signal can be recovered from just the odd/purple blocks or just the even/blue blocks.

Related content
The team’s non-real-time system is the top performer, while its real-time system finishes third overall and second among real-time systems — despite using only 4% of a CPU core.

The degree of redundancy is determined by the number of past chunks included in each packet; each chunk included in the redundancy coding corresponds to 40 ms of speech that can be recovered. Furthermore, rather than representing each chunk independently, the encoder takes advantage of the correlation between successive chunks and extracts a sort of interchunk difference to encode.

For decoding, to be able to synthesize the whole sequence, all we need is a starting point. But rather than decoding forward in time, as would be intuitive, we choose an initial state that corresponds to the most recent chunk; from there, we decode going backward in time. That means we can get quickly to the most recent audio, which is more likely to be useful. It also means that we can transmit as much — or as little — redundancy as we want just by choosing how many chunks to include in a packet.

Rate-distortion-optimized variational autoencoder

Now let's get into the details of how we minimize the bit rate to code our redundancy. Here we turn to a widely used method in the video coding world, rate distortion optimization (RDO), which means trying to simultaneously reduce the bit rate and the distortion we cause to the speech. In a regular autoencoder, we train an encoder to find a simple — typically, low-dimensional — vector representation of an input that can then be decoded back to something close to the original.

In our rate-distortion-optimized variational autoencoder (RDO-VAE), instead of imposing a limit on the dimensionality of the representation, we directly limit the number of bits required to code that representation. We can estimate the actual rate (in bits) required to code the latent representation, assuming entropy coding of a quantized Laplace distribution. As a result, not only do we automatically optimize the feature representation, but the training process automatically discards any useless dimensions by setting them to zero. We don't need to manually choose the number of dimensions.

Moreover, by varying the rate-distortion trade-off, we can train a rate-controllable quantizer. That allows us to use better quality for the most recent speech (which is more likely to be used) and a lower quality for older speech that would be used only for a long burst of loss. In the end, we use an average bit rate of around 500 bits/second (0.5 kb/s) and still have enough information to reconstruct intelligible speech.

Once we include DRED, this is what the packet loss scenario described above would look like:

DRED vs. LBRR.png
With LBRR, each new packet (blue) includes a compressed copy of the previous packet (green); with DRED, it includes highly compressed versions of up to 50 prior packets (red). In this case, DRED's redundancy is set at 140 ms (seven packets).

Although it is illustrated for just 70 milliseconds of redundancy, we scale this up to one full second of redundancy contained in each 20-millisecond packet. That's 50 copies of the information being sent, on the assumption that at least one will make it to its destination and enable reconstruction of the original speech.

Revisiting packet loss concealment

So what happens when we lose a packet and don't have any DRED data for it? We still need to play out something — and ideally not zeros. In that case, we can just guess. Over a short period of time, we can still predict acoustic features reasonably well and then ask LPCNet to fill in the missing audio based on those features. That is essentially what PLC does, and doing it with a neural vocoder like LPCNet works better than using traditional PLC algorithms like the one that's currently integrated into Opus. In fact, our neural PLC algorithm recently placed second in the Interspeech 2022 Audio Deep Packet Loss Concealment Challenge.

Results

How much does DRED improve speech quality and intelligibility under lossy network conditions? Let's start with a clip compressed with Opus wideband at 24 kb/s, plus 16 kb/s of LBRR redundancy (40 kb/s total). This is what we get without loss:

Clean audio

To show what happens in lossy conditions, let's use a particularly difficult — but real — loss sequence taken from the PLC Challenge. If we use the standard Opus redundancy (LBRR) and PLC, the resulting audio is missing large chunks that just cannot be filled:

Lossy audio with LBRR and PLC

If we add our DRED coding with one full second of redundancy included in each packet, at a cost of about 32 kb/s, the missing speech can be entirely recovered:

Lossy audio with DRED
Results.png
Overall results of DRED's evaluation on the full dataset for the original PLC Challenge, using mean opinion score (MOS).

The example above is based on just one speech sequence, but we evaluated DRED on the full dataset for the original PLC Challenge, using mean opinion score (MOS) to aggregate the judgments of human reviewers. The results show that DRED alone (no LBRR) can reduce the impact of packet loss by about half even compared to our previous neural PLC. Also interesting is the fact that LBRR still provides a benefit even when DRED is used. With both LBRR and DRED, the impact of packet loss becomes very small, with just a 0.1 MOS degradation compared to the original, uncompressed speech.

This work is only one example of how Amazon is contributing to improving Opus. Our open-source neural PLC and DRED implementations are available on this development branch, and we welcome feedback and outside collaboration. We are also engaging with the IETF with the goal of updating the Opus standard in a fully compatible way. Our two Internet drafts (draft 1 | draft 2) offer more details on what we are proposing.

Research areas

Related content

US, WA, Bellevue
Join the next science and engineering revolution at Amazon's Delivery Foundation Model team, where you'll work alongside world-class scientists and engineers to pioneer the next frontier of logistics through advanced AI and foundation models. We are seeking an exceptional Senior Applied Scientist to help develop innovative foundation models that enable delivery of billions of packages worldwide. In this role, you'll combine highly technical work with scientific leadership, ensuring the team delivers robust solutions for dynamic real-world environments. Your team will leverage Amazon's vast data and computational resources to tackle ambitious problems across a diverse set of Amazon delivery use cases. Key job responsibilities - Design and implement novel deep learning architectures combining a multitude of modalities, including image, video, and geospatial data. - Solve computational problems to train foundation models on vast amounts of Amazon data and infer at Amazon scale, taking advantage of latest developments in hardware and deep learning libraries. - As a foundation model developer, collaborate with multiple science and engineering teams to help build adaptations that power use cases across Amazon Last Mile deliveries, improving experience and safety of a delivery driver, an Amazon customer, and improving efficiency of Amazon delivery network. - Guide technical direction for specific research initiatives, ensuring robust performance in production environments. - Mentor fellow scientists while maintaining strong individual technical contributions. A day in the life As a member of the Delivery Foundation Model team, you’ll spend your day on the following: - Develop and implement novel foundation model architectures, working hands-on with data and our extensive training and evaluation infrastructure - Guide and support fellow scientists in solving complex technical challenges, from trajectory planning to efficient multi-task learning - Guide and support fellow engineers in building scalable and reusable infra to support model training, evaluation, and inference - Lead focused technical initiatives from conception through deployment, ensuring successful integration with production systems- Drive technical discussions within the team and and key stakeholders - Conduct experiments and prototype new ideas - Mentor team members while maintaining significant hands-on contribution to technical solutions About the team The Delivery Foundation Model team combines ambitious research vision with real-world impact. Our foundation models provide generative reasoning capabilities required to meet the demands of Amazon's global Last Mile delivery network. We leverage Amazon's unparalleled computational infrastructure and extensive datasets to deploy state-of-the-art foundation models to improve the safety, quality, and efficiency of Amazon deliveries. Our work spans the full spectrum of foundation model development, from multimodal training using images, videos, and sensor data, to sophisticated modeling strategies that can handle diverse real-world scenarios. We build everything end to end, from data preparation to model training and evaluation to inference, along with all the tooling needed to understand and analyze model performance. Join us if you're excited about pushing the boundaries of what's possible in logistics, working with world-class scientists and engineers, and seeing your innovations deployed at unprecedented scale.
US, CA, San Diego
We are seeking an exceptional Applied Scientist to join a team of experts in the field of machine learning, and work together to tackle challenging problems across diverse compliance domains. We leverage and train state-of-the-art multi-modal and large-language-models (LLMs) to detect illegal and unsafe products across the Amazon catalog. We work on machine learning problems for multi-modal classification, intent detection, information retrieval, anomaly and fraud detection, and generative AI. This is an exciting and challenging position to deliver scientific innovations into production systems at Amazon-scale to make immediate, meaningful customer impacts while also pursuing ambitious, long-term research. You will work in a highly collaborative environment where you can analyze and process large amounts of image, text and tabular data. You will work on hard science problems that have not been solved before, conduct rapid prototyping to validate your hypothesis, and deploy your algorithmic ideas at scale. There will be something new to learn every day as we work in an environment with rapidly evolving regulations and adversarial actors looking to outwit your best ideas. Key job responsibilities • Design and evaluate state-of-the-art algorithms and approaches in multi-modal classification, large language models (LLMs), intent detection, information retrieval, anomaly and fraud detection, and generative AI • Translate product and CX requirements into measurable science problems and metrics. • Collaborate with product and tech partners and customers to validate hypothesis, drive adoption, and increase business impact • Key author in writing high quality scientific papers in internal and external peer-reviewed conferences. A day in the life - Understanding customer problems, project timelines, and team/project mechanisms - Proposing science formulations and brainstorming ideas with team to solve business problems - Writing code, and running experiments with re-usable science libraries - Reviewing labels and audit results with investigators and operations associates - Sharing science results with science, product and tech partners and customers - Writing science papers for submission to peer-review venues, and reviewing science papers from other scientists in the team. - Contributing to team retrospectives for continuous improvements - Driving science research collaborations and attending study groups with scientists across Amazon About the team We are a team of applied scientists building AI/ML solutions to make Amazon Earth’s most trusted shopping destination for safe and compliant products.
US, CA, San Francisco
Amazon has launched a new research lab in San Francisco to develop foundational capabilities for useful AI agents. We’re enabling practical AI to make our customers more productive, empowered, and fulfilled. In particular, our work combines large language models (LLMs) with reinforcement learning (RL) to solve reasoning, planning, and world modeling in both virtual and physical environments. Our research builds on that of Amazon’s broader AGI organization, which recently introduced Amazon Nova, a new generation of state-of-the-art foundation models (FMs). Key job responsibilities You will contribute directly to AI agent development in an engineering management role: leading a software development team focused on our internal platform for acquiring agentic experience at large scale. You will help set direction, align the team’s goals with the broader lab, mentor team members, recruit great people, and stay technically involved. You will be hired as a Member of Technical Staff. About the team Our lab is a small, talent-dense team with the resources and scale of Amazon. We’re entering an exciting new era where agents can redefine what AI makes possible. We’d love for you to join our lab and build it from the ground up!
US, WA, Seattle
The AWS Supply Chain organization is looking for a Sr. Manager of Applied Science to lead science and data teams working on innovative AI-powered supply chain solutions. As part of the AWS Solutions organization, we have a vision to provide business applications, leveraging Amazon’s unique experience and expertise, that are used by millions of companies worldwide to manage day-to-day operations. We will accomplish this by accelerating our customers’ businesses through delivery of intuitive and differentiated technology solutions that solve enduring business challenges. We blend vision with curiosity and Amazon’s real-world experience to build opinionated, turnkey solutions. Where customers prefer to buy over build, we become their trusted partner with solutions that are no-brainers to buy and easy to use. Are you excited about developing state-of-the-art GenAI/Agentic AI based solutions for enterprise applications? As a Sr. Manager of Applied Scientist at AWS Supply Chain, you will bring AI advancements to customer facing enterprise applications. In this role, you will drive the technical vision and strategy for your team while fostering a culture of innovation and scientific excellence. You will be leading a fast-paced, cross-disciplinary team of researchers who are leaders in the field. You will take on challenging problems, distill real requirements, and then deliver solutions that either leverage existing academic and industrial research, or utilize your own out-of-the-box pragmatic thinking. In addition to coming up with novel solutions and prototypes, you may even need to deliver these to production in customer facing products. Key job responsibilities Building and mentoring teams of Applied Scientists, ML Engineers, and Data Scientists. Setting technical direction and research strategy aligned with business goals. Driving innovation in Supply Chains systems using AI/ML models and AI Agents. Collaborating with cross-functional teams to translate research into production. Managing project portfolios and resource allocation.
US, NY, New York
Are you a passionate Applied Scientist (AS) ready to shape the future of digital content creation? At Amazon, we're building Earth's most desired destination for creators to monetize their unique skills, inspire the next generation of customers, and help brands expand their reach. We build innovative products and experiences that drive growth for creators across Amazon's ecosystem. Our team owns the entire Creator product suite, ensuring a cohesive experience, optimizing compensation structures, and launching features that help creators achieve both monetary and non-monetary goals. Key job responsibilities As an AS on our team, you will: - Handle challenging problems that directly impact millions of creators and customers - Independently collect and analyze data - Develop and deliver scalable predictive models, using any necessary programming, machine learning, and statistical analysis software - Collaborate with other scientists, engineers, product managers, and business teams to creatively solve problems, measure and estimate risks, and constructively critique peer research - Consult with engineering teams to design data and modeling pipelines which successfully interface with new and existing software - Participate in design and implementation across teams to contribute to initiatives and develop optimal solutions that benefit the creators organization The successful candidate is a self-starter, comfortable with a dynamic, fast-paced environment, and able to think big while paying careful attention to detail. You have deep knowledge of an area/multiple areas of science, with a track record of applying this knowledge to deliver science solutions in a business setting and a demonstrated ability to operate at scale. You excel in a culture of invention and collaboration.
GB, MLN, Edinburgh
Do you want a role with deep meaning and the ability to make a major impact? As part of Intelligent Talent Acquisition (ITA), you'll have the opportunity to reinvent the hiring process and deliver unprecedented scale, sophistication, and accuracy for Amazon Talent Acquisition operations. ITA is an industry-leading people science and technology organization made up of scientists, engineers, analysts, product professionals and more, all with the shared goal of connecting the right people to the right jobs in a way that is fair and precise. Last year we delivered over 6 million online candidate assessments, and helped Amazon deliver billions of packages around the world by making it possible to hire hundreds of thousands of workers in the right quantity, at the right location and at exactly the right time. You’ll work on state-of-the-art research, advanced software tools, new AI systems, and machine learning algorithms, leveraging Amazon's in-house tech stack to bring innovative solutions to life. Join ITA in using technologies to transform the hiring landscape and make a meaningful difference in people's lives. Together, we can solve the world's toughest hiring problems. Key job responsibilities As an Applied Scientist, you will own the design and development of end-to-end systems. You’ll have the opportunity to write technical white papers, create technical roadmaps and drive production level projects that will support Amazon Science. You will have the opportunity to design new algorithms, models, or other technical solutions whilst experiencing Amazon’s customer focused culture. The ideal scientist must have the ability to work with diverse groups of people and cross-functional teams to solve complex business problems. About the team The Automated Performance Evaluation (APE) team is a hybrid team of Applied Scientists and Software Development Engineers who develop, deploy and own end-to-end machine learning services for use in the HR and Recruiting functions at Amazon.
US, CA, Sunnyvale
Prime Video is a first-stop entertainment destination offering customers a vast collection of premium programming in one app available across thousands of devices. Prime members can customize their viewing experience and find their favorite movies, series, documentaries, and live sports – including Amazon MGM Studios-produced series and movies; licensed fan favorites; and programming from Prime Video subscriptions such as Apple TV+, HBO Max, Peacock, Crunchyroll and MGM+. All customers, regardless of whether they have a Prime membership or not, can rent or buy titles via the Prime Video Store, and can enjoy even more content for free with ads. Are you interested in shaping the future of entertainment? Prime Video's technology teams are creating best-in-class digital video experience. As a Prime Video team member, you’ll have end-to-end ownership of the product, user experience, design, and technology required to deliver state-of-the-art experiences for our customers. You’ll get to work on projects that are fast-paced, challenging, and varied. You’ll also be able to experiment with new possibilities, take risks, and collaborate with remarkable people. We’ll look for you to bring your diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. With global opportunities for talented technologists, you can decide where a career Prime Video Tech takes you! Key job responsibilities As an Applied Scientist at Prime Video, you will have end-to-end ownership of the product, related research and experimentation, applying advanced machine learning techniques in computer vision (CV), Generative AI, multimedia understanding and so on. You’ll work on diverse projects that enhance Prime Video’s content localization, image/video understanding, and content personalization, driving impactful innovations for our global audience. Other responsibilities include: - Research and develop generative models for controllable synthesis across images, video, vector graphics, and multimedia - Innovate in advanced diffusion and flow-based methods (e.g., inverse flow matching, parameter efficient training, guided sampling, test-time adaptation) to improve efficiency, controllability, and scalability. - Advance visual grounding, depth and 3D estimation, segmentation, and matting for integration into pre-visualization, compositing, VFX, and post-production pipelines. - Design multimodal GenAI workflows including visual-language model tooling, structured prompt orchestration, agentic pipelines. A day in the life Prime Video is pioneering the use of Generative AI to empower the next generation of creatives. Our mission is to make world-class media creation accessible, scalable, and efficient. We are seeking an Applied Scientist to advance the state of the art in Generative AI and to deliver these innovations as production-ready systems at Amazon scale. Your work will give creators unprecedented freedom and control while driving new efficiencies across Prime Video’s global content and marketing pipelines. This is a newly formed team within Prime Video Science!
IN, KA, Bengaluru
Amazon Devices is an inventive research and development company that designs and engineer high-profile devices like the Kindle family of products, Fire Tablets, Fire TV, Health Wellness, Amazon Echo & Astro products. This is an exciting opportunity to join Amazon in developing state-of-the-art techniques that bring Gen AI on edge for our consumer products. We are looking for exceptional early career research scientists to join our Applied Science team and help develop the next generation of edge models, and optimize them while doing co-designed with custom ML HW based on a revolutionary architecture. Work hard. Have Fun. Make History. Key job responsibilities Key Job Responsibilities: • Understand and contribute to model compression techniques (quantization, pruning, distillation, etc.) while developing theoretical understanding of Information Theory and Deep Learning fundamentals • Work with senior researchers to optimize Gen AI models for edge platforms using Amazon's Neural Edge Engine • Study and apply first principles of Information Theory, Scientific Computing, and Non-Equilibrium Thermodynamics to model optimization problems • Assist in research projects involving custom Gen AI model development, aiming to improve SOTA under mentorship • Co-author research papers for top-tier conferences (NeurIPS, ICLR, MLSys) and present at internal research meetings • Collaborate with compiler engineers, Applied Scientists, and Hardware Architects while learning about production ML systems • Participate in reading groups and research discussions to build expertise in efficient AI and edge computing
US, NY, New York
The Artificial General Intelligence (AGI) team is looking for a passionate, talented, and inventive Applied Scientist to work on pre-training methodologies for Generative Artificial Intelligence (GenAI) models. You will interact closely with our customers and with the academic and research communities. Key job responsibilities Join us to work as an integral part of a team that has experience with GenAI models in this space. We work on these areas: - Scaling laws - Hardware-informed efficient model architecture, low-precision training - Optimization methods, learning objectives, curriculum design - Deep learning theories on efficient hyperparameter search and self-supervised learning - Learning objectives and reinforcement learning methods - Distributed training methods and solutions - AI-assisted research About the team The AGI team has a mission to push the envelope in GenAI with Large Language Models (LLMs) and multimodal systems, in order to provide the best-possible experience for our customers.
US, CA, Sunnyvale
The Artificial General Intelligence (AGI) team is looking for a passionate, talented, and inventive Applied Scientist with a strong deep learning background, to build industry-leading Generative Artificial Intelligence (GenAI) technology with Large Language Models (LLMs) and multimodal systems. Key job responsibilities As an Applied Scientist with the AGI team, you will work with talented peers to support the development of algorithms and modeling techniques, to advance the state of the art with LLMs. Your work will directly impact our customers in the form of products and services that make use of GenAI technology. You will leverage Amazon’s heterogeneous data sources and large-scale computing resources to accelerate advances in LLMs. About the team The AGI team has a mission to push the envelope in GenAI with LLMs and multimodal systems, in order to provide the best-possible experience for our customers.