Following are a few white papers. You can download them from the “Latest Thinking” section of my blog: https://asiliconvalleyinsider.com/latest-thinking/ or from the links below.

Language models (LLs) have been through three main evolutions: Word2vec, Recurrent Neural Networks (RNNs)/Long Short-Term Memory Networks (LSTM), and more recently Transformers architectures. Transformer networks have enabled to design LLs for natural language understanding (NLP) applications that can scale to very large data sets. As a result, LLs became large language models (LLMs). And, LLMs such as OpenAI’s GPT-3 autoregressive model brought new capabilities to NLP applications as they can learn, after being pre-trained, a new task using zero-shot or a few shots learning, that is with no or only a few examples and without any fine-tuning that is no model training parameter update, a capability called in-context learning that we still cannot explain.

Read more: Large Language Models

The goal of Machine Learning (ML) is to develop algorithms for making predictions from an existing data set. First, you start by collecting the data and analyzing it. Second, you build a model for it and select an ML algorithm. Third, you evaluate and tune the model and the algorithm. Last, you deploy it and make live predictions. There are three major approaches to ML: Regression finds a function to which a new instance belongs to; Classification establishes the category (or class) to which a new instance belongs; And clustering groups/clusters the instances into categories (or classes).

Read more: Machine Learning

The Internet is running out of addresses! Depending on how long the present pool of IPv4 addresses can be preserved, the exhaustion could begin as soon as Spring 2011. IPv6 is the “only” solution to the exhaustion of Internet addresses. Unfortunately, there is little worldwide network deployment of IPv6. And, without any migration from IPv4 to IPv6, the Internet will run out of addresses. IPv6 introduces a number of features and benefits to the Internet Protocol for mobility, routing, QoS and security.

Read more: IPv6

IP packets have no inherent security. To that end, IPSec was designed to provide network-layer security services for IP upper-layer protocols (TCP and UDP). It guarantees that the message actually was sent by the apparent originator of the message; that the message that is received is the exact one that was sent; that the message cannot be read by a third party.

Read more: IPSecurity

While IPSec secures IP traffic, the management and scale of security policies and the encryption key exchange mechanisms using IKE create a number of practical limitations. By dividing policy and key generation into separate logical components and combining them in new ways, a network security overlay can be deployed to protect data from end-to-end without any impact on the network architecture and performance.

Read more: Overlay Network for Security Policies

VoIP needs to be encoded and decoded. To that end, many codecs have been proposed by vendors and standardized. VoIP quality is affected both by the conditions of the network and the user at each end-point.

Read more: Voice Encoding

Organizations have been demanding virtual private networks (VPNs) instead of costly leased lines and, Ethernet and IP services instead of DS1 and DS3 services from their service providers. MPLS is now the best way for service providers to respond to those customer demands by providing Pseudo Wires (PW) services, Virtual Private LAN Services (VPLS) or Layer 2 MPLS VPNs and IP VPNs or Layer 3 MPLS VPNs.

Read more: MPLS VPN Services

For existing access networks, Ethernet can be transported over SONET (EoS) using three new emerging technologies: the Generic Framing Protocol (GFP), Virtual Concatenation (VCAT) and Link Capacity Adjustment Scheme (LCAS). GFP defines an Ethernet client signal mapping into SONET payloads. VCAT and LCAS provide the concatenation and the sizing of the SONET frames. The result is a much more efficient use of the SONET bandwidth.

Read more: Ethernet Transport

Released in 1996, the Java Telephony (JTAPI) was the first application programming interface to support both first-party (desktop telephony) and third-party (server telephony) applications. The core of the API provides a basic call control model for rudimentary telephony features such as Internet telephony. The standard extensions of the API provide advanced functionality for various telephony capabilities such as media streams or telephony applications such as call centers.

Read more: The Java Telephony API